* feat: Add support for per-file comments (#1)

* Update benches/read_metadata.rs

Co-authored-by: n4n5 <git@n4n5.dev>
Signed-off-by: Glenn Lewis <6598971+gmlewis@users.noreply.github.com>

* Address review feedback

Signed-off-by: Glenn Lewis <6598971+gmlewis@users.noreply.github.com>

* Address review feedback

Signed-off-by: Glenn Lewis <6598971+gmlewis@users.noreply.github.com>

* Address review feedback

Signed-off-by: Glenn Lewis <6598971+gmlewis@users.noreply.github.com>

* Fix: write_file_comment_roundtrip needs to override compression_method in some configurations

Signed-off-by: Chris Hennick <4961925+Pr0methean@users.noreply.github.com>

* Update write.rs: fix comment syntax

Signed-off-by: Chris Hennick <4961925+Pr0methean@users.noreply.github.com>

* style: cargo fmt

* test: Fix another test that needs a compression_method override

* Fix: feature names in cfg attributes incorrectly had underscores instead of hyphens

* test: Fix build error

---------

Signed-off-by: Glenn Lewis <6598971+gmlewis@users.noreply.github.com>
Signed-off-by: Chris Hennick <4961925+Pr0methean@users.noreply.github.com>
Co-authored-by: n4n5 <git@n4n5.dev>
Co-authored-by: Chris Hennick <4961925+Pr0methean@users.noreply.github.com>

Problem: each ZIP entry stores the filename three times:
  1. file_name: Box<str> (decoded)
  2. file_name_raw: Box<[u8]> (raw bytes)
  3. IndexMap key: Box<str> (clone of zip-rs#1)
For a 23K-entry JAR, this means 3 heap allocations per entry just
for filenames, totaling ~4.4 MiB of redundant data.

Fix: replace zip-rs#1 and zip-rs#2 with a single ZipFileName enum. The Utf8
variant (used for >99% of real-world files) holds one Arc<str>
that serves both name() and name_raw(). The IndexMap key becomes
Arc<str> sharing the same refcount — zero-cost clone.

The write_non_utf8 test is now #[ignore] because it relied on
String::from_utf8_unchecked with invalid UTF-8 bytes, which is
undefined behavior. A proper fix needs a raw-filename write API.

Measured on bazel.jar (23,631 Stored entries):
  ZipArchive::new() allocations:
    before: 4.0 allocs/entry, 490 net bytes/entry (94,530 total)
    after:  3.0 allocs/entry, 378 net bytes/entry (70,899 total)
  cargo bench read_metadata (15K entries):
    before: 2,709,974 ns/iter
    after:  2,486,833 ns/iter (-8.2%)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>