Optimize library installer conan calls

Now that the Cloudfront caching in front of the Conan server is disabled, we need to figure out a way to manually optimize the process.

A large IO bottleneck is the checks that we do to prevent from building the same thing over and over again:
* `has_failed_before` https://github.com/compiler-explorer/infra/blob/main/bin/lib/library_builder.py#L783
* `is_already_uploaded` https://github.com/compiler-explorer/infra/blob/main/bin/lib/library_builder.py#L796
- `get_build_annotations` https://github.com/compiler-explorer/infra/blob/main/bin/lib/library_builder.py#L752 is part of this

These are called for every compiler that will be built with. There might be a way to request the information for all compilers at once and then only do the lookup per compiler.

Related endpoint: https://github.com/compiler-explorer/conanproxy/blob/main/index.js#L454
* See also: https://github.com/compiler-explorer/conanproxy/blob/main/build-annotations.js#L18
* Unfortunately the conan_hash will be unique per compiler, but if you iterate over all the directories in the package directory (https://github.com/compiler-explorer/conanproxy/blob/main/build-annotations.js#L11C102-L11C111) you can match the hashes on the builder side

Lesser influence, but still IO:
* `get_commit_hash` https://github.com/compiler-explorer/infra/blob/85647382536eeb1284845497b94850cb8c8427be/bin/lib/library_builder.py#L769
- This really only has to be called once, so either the first time its called or just in the constructor

Start in bin/lib/library_builder.py by reading has_failed_before, is_already_uploaded, get_build_annotations, and get_commit_hash. Compare these calls with the conanproxy endpoint in index.js and build-annotations.js; done means reducing repeated IO while preserving the per-compiler hash matching and calling get_commit_hash only once.

Written by the indexing model from the issue text.