Fix raidfile tests on 32-bit Linux.

A recent fix for Solaris (commit 81e9aa65) broke support for 32-bit Linux (which wasn't spotted at the time, because we didn't have any 32-bit builders). Try a different approach: detect explicitly whether the `lseek` syscall takes a 64-bit integer offset, regardless of the size of `off_t` in user space.
    
CMake: Add support for M4 CXX flag detection in CMakeLists. Reimplement autoconf tests for 64-bit lseek.

Fix error in t-gdb when no debugger is detected.

Thanks to Reinhard Tartler (our Debian package maintainer) for pointing out the error, and James O'Gorman for setting up i386 builders to ensure that it's fixed and cannot recur.

Enables all flags that are detected by M4 builds to be detected by CMake builds
as well (except for -Wall which enables a bunch of junk).

(cherry picked from commit 2ea8a970)
(cherry picked from commit c6038419)

A recent fix for Solaris (commit 81e9aa65)
broke support for 32-bit Linux (which wasn't spotted because we didn't have any
32-bit builders). Try a different approach: detect whether the lseek syscall
takes a 64-bit integer offset, and use that if possible.

CMake: reimplement autoconf tests for 64-bit lseek

(cherry picked from commit 138ea5d174f146f14d91a16bf5d1ce8e479d2024)

Fix OpenSSL 1.1 compatibility

Download NSIS and Boost archives ourselves and install in correct places for
Chocolatey and CMake to find and use them, instead of trying to download them
from SourceForge.

See https://github.com/alex/pyopenssl/commit/02225c659a1441e5caeca0a327f85b241633f781
for example

This time I have made the minimum number of changes necessary to make OpenSSL
1.1 work, and also added more tests which would have caught the issues with the
previous attempt.

Thanks to Johann Glaser for testing out the new version and reporting the
issue with decryption!

Until shown to be safe. See message from Johann Glaser to the mailing list.

This reverts commit 85e7efc3.

Fix compatibility with OpenSSL 1.1

See https://github.com/boxbackup/boxbackup/issues/16 for details. Thanks to
Chris West @FauxFaux for the initial patch!

Use OpenSSL 1.1.0g for Windows superbuild.

Fix Windows build by removing calls to obsolete do_ms.bat
<https://stackoverflow.com/questions/39076244/why-there-is-no-ms-do-ms-bat-after-perl-configure-vc-win64a/39247560#39247560>.

Workaround for incorrect library suffixes searched by FindOpenSSL
<https://gitlab.kitware.com/cmake/cmake/issues/17604>.

Link OpenSSL statically and fix missing dependency on crypt32.lib.

(cherry picked from commit edd3687f067c68b131822e0064cdeff5bf7a3835)
(cherry picked from commit b003e009d1fccadf55a0f59a01c66f7b80f9b3e5)
(cherry picked from commit cc6e20466b857f7e20d816a2c20cb36f112398ab)
(cherry picked from commit 303c6405e3bc1e673dbc691792e0633742c01f4e)
(cherry picked from commit 4e24006f75142a254f1a7db76cf9da1e752d9283)
(cherry picked from commit 448ac48a89538f67f508f8d21931c5e6e90472c0)
(cherry picked from commit 00b3bb0697219ef8c742698dd43d53b538e362b1)

RAND_pseudo_bytes is less secure and is now deprecated.

(cherry picked from commit f449986b68bd12c18d742c91b625779ae75b11a2)

Previously, it was impossible to include an @ sign in these flags (for example,
to link to support --with-ssl-headers=/usr/local/opt/openssl@1.1/include/)
because it was interpreted as a list variable name by Perl. Now the variables
are evaluated single-quoted before being substituted into the Makefile output.

(cherry picked from commit 2e9c90d3f32c69cc89d01fb64ced9fca80aae985)

Should make debugging on Mac and BSD less painful.

Make t-gdb fail with an error when no debugger was detected.

(cherry picked from commit da98a818)
(cherry picked from commit ea6bc722)
(cherry picked from commit 38ba7510)
(cherry picked from commit 02bfd7e5bdd2abc185b12f37e1f536269db6817a)

We don't need to install anything in C:\Program Files\BoxBackup, and often
don't have permission to do that anyway.

(cherry picked from commit 2d31a317)
(cherry picked from commit dab8ccc4b0b9eef06c9e063fc510f9952a0757f7)

(cherry picked from commit 8976906a563ab04eb3d1310c59d4c13cdc3b525e)

Build Win32/64 binary packages automatically

Enable the relevant compiler warning as an error, if supported
(-Werror=undefined-bool-conversion).

http://www.gotw.ca/conv/002.htm
http://stackoverflow.com/questions/2165078/a-reference-can-not-be-null-or-it-can-be-null

(cherry picked from commit f2911aca)
(cherry picked from commit 5a50b984)

Fix parsing of OpenSSL 1.1 output in bbstored-certs. Fixes #18. Thanks to Dean Hamstead (@djzort on GitHub) for the bug report and the patch!

Master build fails with: lib/common/Timer.cpp:171:10: error: reference cannot
be bound to dereferenced null pointer in well-defined C++ code; pointer may be
assumed to always convert to true [-Werror,-Wundefined-bool-conversion].

Cannot get regtests to pass on this feature branch without fixing this.

Also remove unused code from Timers.

Thanks to Dean Hamstead (@djzort on GitHub) for the bug report and the patch!

Otherwise the unibuild fails at the end for non-administrator users when
it tries to write to C:\Program Files\BoxBackup.

(cherry picked from commit 793d5f57f5f99f50bffadea674f58be79f9cb6ff)

(cherry picked from commit 4876a5573a8bcd3312e9804c27fe9f30f3369747)

(cherry picked from commit ea4f1728b529aff1dfc912d8f51567f8cd58ea77)

See
http://help.appveyor.com/discussions/problems/4569-the-target-_convertpdbfiles-listed-in-a-beforetargets-attribute-at-c-does-not-exist-in-the-project-and-will-be-ignored for details.

(cherry picked from commit 019025deceff49485b03593f7a77a570d960546a)

Every AppVeyor build on the specified branches that passes will be packaged
into a ZIP file and installer using CPack and uploaded to GitHub Releases.

Generate version number entirely in AppVeyor config, and build it into the
binary, and use it in the names of the created binary packages, and compile it
into the binaries.

(cherry picked from commit c5d8173bf03ab9fa93d3b3ffd871b378b0819567)

Fix build on ARM64 with unsigned char

Thanks to Reinhard Tartler for pointing out the problem and suggesting the fix!

Should hopefully reduce random failures of timing-dependent tests which
rely on subsecond precision for accuracy.

(cherry picked from commit c845b2e3)

Ideally timers would be perfectly accurate and we could sleep for 1.0 seconds,
but on OSX in particular they could fire 50-100 ms late (I've seen 4 ms in
practice) and we don't want the tests to fail because of this, because we don't
really need that kind of precision in practice. So we reduce the timer
intervals by 100ms to be safe.

Increase logging level and add timestamps in testcommon.  Should help to debug
frequent test failures such as Condition [t2.HasExpired()] on OSX hosts on
Travis.

Add a SettingsGuard to the Console logger for its specific settings, such as
showing times and microseconds.

(cherry picked from commit a5e6cff1d435329b0121417ed9509e315ce0edd5)
(cherry picked from commit 8d02eebc)