FileID needs some work on Linux

Would be good to bring it up to parity with the Windows implementation, using something like GNU .build-id if it's available, and falling back to some objective measure of the file itself (like a hash of its timestamp/size/base name) that doesn't include a hash of its full path.

Right now the ELF parsing we do happens after the FileID is generated, so even if .build-id parsing was added there we likely need to reorganize things a bit.

Start by tracing the current FileID generation and the existing ELF parsing path described in the issue, then compare the Linux and Windows implementations. Check how GNU .build-id availability can be detected and define a fallback based on file metadata rather than its full path; done means Linux FileID behavior reaches parity without relying on post-generation ELF parsing.

Written by the indexing model from the issue text.