* perf: skip BufReader for Stored files in make_reader

Problem: by_index() allocates an 8 KiB BufReader and a Decompressor
for every entry, even CompressionMethod::Stored where decompression
is a no-op pass-through. For a 23K-entry JAR (mostly Stored files),
this means 2 heap allocations per call totaling 8,632 bytes each —
47,262 allocations just to iterate and drop.

Fix: add a ZipFileReader::Stored variant that wraps CryptoReader
directly in Crc32Reader, bypassing both BufReader and Decompressor.

This required adding a Debug impl for CryptoReader since the new
Stored variant needs it for the derive(Debug) on ZipFileReader.

Measured on bazel.jar (23,631 Stored entries):
  by_index() allocations:
    before: 2.0 allocs/entry, 8,632 bytes alloc'd each
    after:  1.0 allocs/entry,   256 bytes alloc'd each
  cargo bench read_entry (1 MiB Stored file):
    before: 106,657 ns/iter (9,831 MB/s)
    after:  101,276 ns/iter (10,353 MB/s, +5.3%)

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

* move Debug defition after type and add debug information for Aes

* cargo fmt

---------

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>
Co-authored-by: n4n5 <[email protected]>

* refactor: part of read.rs

* add reader mod

* move to zip_archive

* move

* re-add imports

* move tests

* move again

* revert fmt

* fmt

* fix tests

* move aes info

* move function

* move from reader.rs to readers.rs

* self contained test

* move more reader to readers

* typos

* fmt and fix CI

* use macros

* cargo fmt

* cargo fmt

* chore: move template to correct place

* fix: edit template

* feat: add a check for building benches

* Trigger CI