In order to parse the annotated output it would be nice if that would contain
information about the prefix (format string) it uses.

This is done by printing another line similar to the “header”, which gives the
version information, thereby again altering the output format.

Tools, that parse the annotated output should in principle assume that any lines
`I: …` without a prefix may be included in the output (in principle at any
point).

In order to no need a shell variable (which might unintentionally get passed on
to the actually executed program, see commit
0d5ca640 (“annotate-output: prevent `PROGNAME`-
variable from leaking to command”)), the construct with a definition function is
used again.

The double-escaping using GNU `printf`’s `%q`-conversion-specifier is necessary
because once it’s needed for the `eval` and once to have it escaped as output.

The leading `+`, if any, is stripped off because it’s not really part of the
prefix format string, but only needed for the `date`-utility.

The `get_prefix()`-function could have been defined to always print a trailing
`\n`, thereby avoiding the need for the final `printf '\n'`, but the, the
function – should it ever be used for some further purpose – would not return
the real prefix.

Signed-off-by: Christoph Anton Mitterer <mail@christoph.anton.mitterer.name>