Inspect GNU thin archives during ELF ABI validation

## Current evidence

`src/link_validation/elf.rs` recognizes direct ELF files and ordinary Unix archives beginning with `!\n`. Other inputs are intentionally ignored as formats that do not carry inspectable ELF metadata.

GNU thin archives use the separate `!\n` format and reference member files instead of embedding each member payload. They currently fall through as an unrecognized input, so RISC-V and LoongArch pre-link ABI validation can skip incompatible ELF members stored through a thin archive.

## Scope

- Recognize GNU thin archives explicitly.
- Resolve referenced members relative to the archive location according to the thin-archive format.
- Inspect ELF member machine and `e_flags` through the existing validation path.
- Preserve handling of ordinary GNU/BSD archives, direct ELF objects, linker scripts, and LLVM bitcode.
- Reject malformed thin-archive metadata with a bounded diagnostic instead of following invalid offsets or paths blindly.

## Completion criteria

- Fixture tests cover a valid thin archive, a mixed/mismatched RISC-V or LoongArch member, long member names, and malformed/truncated metadata.
- ABI mismatches inside thin archives are rejected before the linker is invoked.
- Existing ordinary archive validation continues to pass.

Start in src/link_validation/elf.rs and trace how direct ELF files and ordinary archives reach the existing ABI validation path. Read the thin-archive format details before adding coverage for valid, mixed or mismatched, long-name, malformed, and truncated fixtures. Done means thin-archive members are validated safely while ordinary archives, direct ELF objects, linker scripts, and LLVM bitcode retain their current behavior.

Written by the indexing model from the issue text.