Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

* cli: Add missing return type annotations

* bdist_wheel: Remove dead code for postprocessing WHEEL line endings

The default policies of Message and BytesGenerator always produce LF
line endings, so there's never a CRLF to match.  If there were, the
existing code would have incorrectly replaced CRLF with CR, and then
test_wheelfile_line_endings() would have failed.

* cli: Fix removing build tag with `wheel pack --build-number ""`

There's an existing test for this case, which verifies that the original
build tag is removed from the wheel filename and is *not* removed from
the WHEEL metadata.  Presumably this is not what was intended.

BUILD_NUM_RE assumed that it would only match at end of file, but there's
at least a newline in the way.  Fix it, and the test.

* cli: Allow removing build tag with `wheel tags --build ""`

For symmetry with `wheel pack`.  The functionality works already; we just
need to fix the command-line validation.

* test: Restructure test_pack() around email.message.Message

We're about to change pack() to emit WHEEL metadata with LF-terminated
lines and a trailing blank line.  In some test_pack() parameterizations
we expect to see the original WHEEL file from the test fixture, and in
some, a modified one.  Handle the difference by comparing WHEEL contents
parsed with email.parser, instead of byte strings.

* cli: Remove hand-rolled WHEEL parsing/mutation code

The PyPA Binary Distribution Format spec says the WHEEL metadata file is
"in the same basic key: value format" as the METADATA file.  METADATA in
turn is governed by the Core Metadata Specifications, which say: "In the
absence of a precise definition, the practical standard is set by what the
standard library email.parser module can parse using the compat32 policy."

wheel.bdist_wheel accordingly uses email.generator.BytesGenerator to
generate WHEEL, but the CLI `pack` and `tags` commands opt to
hand-implement WHEEL parsing and mutation.  Their mutation functions,
set_tags() and set_build_number(), append any new headers to the existing
WHEEL file.  Since WHEEL tends to have a trailing blank line (separating
the headers from the nonexistent body), the new headers end up in the
"body" and are ignored by any tool that parses WHEEL with email.parser.

In addition, both functions assume that WHEEL uses CRLF line endings,
while bdist_wheel (and email.generator with the compat32 policy)
specifically always writes LF line endings.  This turns out okay for
set_tags(), which rewrites the whole file, but set_build_number() ends up
appending a CRLF-terminated line to a file with LF-terminated lines.

Fix all this by removing the hand-written parsing/mutation functions in
favor of email.parser and email.generator.