Remove strict initialism code by @noctuid

The function `orderless--separated-by` has also been added by @noctuid. However
with the removal of the strict initialism code, the contributions by @noctuid
are pushed below the 15 lines limit. The function `orderless--separated-by` has
only 11 lines.

(cl-defun orderless--separated-by (sep rxs &optional (before "") (after ""))
  "Return a regexp to match the rx-regexps RXS with SEP in between.
If BEFORE is specified, add it to the beginning of the rx sequence. If AFTER is
specified, add it to the end of the rx sequence."
  (rx-to-string
   `(seq
     ,before
     ,@(cl-loop for (sexp . more) on rxs
                collect `(group ,sexp)
                when more collect `,sep)
     ,after)))

Fix orderless-try-completion for hash tables and alists

Optimize orderless-try-completion

If orderless finds that some component of the pattern compiles to a
regexp of the form ^literal it will now take the first such regexp and
add the literal to the prefix which which it calls all-completions.

That's a slight lie: it actually looks for regexps of the form
\(?:^literal\) since that is what the pattern compiler actually
produces.

With this change people can now add a style dispatcher that adds a ^
to the first component to use with certain completion-at-point
functions that can really take advantage of knowing a prefix to return
fewer completions. Some even (incorrectly) refuse to return all
possible completions for an empty string!

See the discussion in #79 and the issues linked therein.

orderless-escapable-split-on-space: Fix escaping mistake

My patch 268e56d4 was not correct.

Fix orderless-escapable-split-on-space

* Currently escaped backslash also escapes space
* Allocate match-data only once

In the common case where people search for a string with no special
regexp characters and are using both the orderless-regexp and
orderless-literal matching styles, a regexp containing blah\|blah was
produced. This takes care of that case.

Use minibuffer customization group

The completion customization group is defined by completion.el,
which is unrelated to minibuffer completion.

I think it's probably less surprising to use the immutable [:alpha:] rather than [:word:] which is up to the current syntax table.

Use add-face-text-property

README: Add variable category to orderless+initialism sample

Marginalia provides the command, symbol and variable categories.

Only enable regexp and literal matching styles by default

* Avoid performance issues with initialism when filtering long candidates (#39)
* README: Update default matching styles
* README: Document `orderless-define-completion-style`
* README: Update `without-if-bang` such that it ignores "!"

fix typo

Remove pattern compiler variable