fix(cache): include subdir in CacheKey to prevent cross-platform cache collisions

[ritankarsaha]

Description

Fixes the long-standing TODO in cache_key.rs about non-uniqueness of CacheKey across subdirectories.

Problem

CacheKey was keyed only on name + version + build_string. Two packages from different platforms
(e.g. linux-64 and osx-arm64) with identical coordinates would hash to the same cache entry. The
first one cached would be returned for both, silently installing the wrong architecture.

Solution

• Added subdir: Option to CacheKey and BucketKey
• From<&PackageRecord> now populates subdir from record.subdir
• Display includes subdir in the directory name when no origin_hash is present:
  • Before: zlib-1.3.1-hb9d3cd8_2
  • After: zlib-1.3.1-hb9d3cd8_2-linux-64
• When origin_hash is set (URL-based fetch), it takes precedence — the URL already encodes the
  subdir, so behavior is unchanged for that path
• Added with_subdir() builder for keys created from CondaArchiveIdentifier (filename-only, no
  subdir info) where the caller knows the platform

Fixes #2166

How Has This Been Tested?

6 unit tests has been added to this properly

The critical test is test_cross_platform_keys_are_distinct — it directly exercises the bug scenario
(two packages from linux-64 vs osx-arm64 with identical coordinates) and asserts they produce different cache directory names.

To run just these tests:
pixi run -- cargo nextest run -p rattler_cache cache_key::tests

Checklist:

• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• I have added sufficient tests to cover my changes.

Breaking change

Cache directory names for PackageRecord-derived keys will include the subdir suffix. Existing
cached entries without the suffix will be treated as misses and re-fetched. This is the correct and
safe behavior.

[baszalmstra]

Did you just open the same pr twice with a different description?

Edit: nope it was two different people.

[ritankarsaha]

Did you just open the same pr twice with a different description?

Ahh no, I squashed the 3 commits into one, to make it look cleaner, that's all !

[baszalmstra]

This is a duplicate of #2165, please see my comment there and figure out amongst yourselves which PR should be merged.

[ritankarsaha]

This is a duplicate of #2165, please see my comment there and figure out amongst yourselves which PR should be merged.

Ahh I see the PR there, but I think you have left a comment there mentioning that #2165 is incomplete and doesn't fix the issue, also both addresses different issue tags.

[baszalmstra]

Both PRs suffer from the exact same issue, this one just has more unit tests. But both try to solve the same problem.

[ritankarsaha]

Both PRs suffer from the exact same issue, this one just has more unit tests. But both try to solve the same problem.

Ahh lemme know, which one you planning to move on with, happy to be of help and use in both @baszalmstra

[baszalmstra]

I dont really care. Maybe you can discuss with the other author how you'd solve my comment?