dat: validate archive inputs and unit completeness

[itsgrimetime]

Reject missing or empty DAT archive directories during CMake configuration instead of successfully producing an empty objdiff project. Report the input path and extraction guidance.

Mark a sample unit complete only when every target symbol outside its samples is also inferred in the base, alongside the existing public-symbol, relocation, and walk checks. Check the symbol count first, then each symbol's file location and size, before inspecting the archive walks. A typed symbol start can still leave unexplained trailing data and a missing base definition.

Add a regression check against real generated objects, comparing inferred symbols by name and size. It uses MELEE_DAT_BUILD or the local DAT preset output, skipping the default when absent. No synthetic DATs or generated artifacts are committed. Document both behaviors.

Validation:

• The real dat preset builds successfully with Clang 22.1.7 and LLD 23.1.1: all 838 archives, 2,201 generated C files, 838 base objects, and 765 objdiff units. The generated project opens in the GUI.
• The regression check fails before the fix on missing DbCo tables. After rebuilding, 246 incorrect complete flags become false; all 6,327 inferred symbols in the remaining 37 complete units pass.
• Cargo tests pass against real archives and the generated DWARF ELF. Only the existing root_witnesses test skips because it hardcodes an object under the separate ppc-dwarf preset directory.
• samples report succeeds with objdiff-cli 3.8.2: 11,945/11,947 samples match, covering 99.9984% of sampled bytes; no extra base sections. The two existing sample mismatches are outside this change's scope.
• Missing, empty, non-archive-only, and file-valued input paths fail configuration. Existing Ninja builds also fail on reconfiguration.
• Clippy completes with three existing warnings. The new test's formatting, git diff --check, and commit hooks pass.

Human review is pending.

[ribbanya]

You might want to hold off on this. I'm in the process of swapping to full cmake-ninja in master...ribbanya:melee:wip/dat.

[ribbanya]

And I don't want to introduce synthetic DATs at this time.

[ribbanya]

@itsgrimetime Hit my session limit and merged in #3590; feel free to continue on with what's still relevant.

[itsgrimetime]

@itsgrimetime Hit my session limit and merged in #3590; feel free to continue on with what's still relevant.

Pivoted this one to work off that

[ribbanya]

Nice, this makes way more sense to look at. Later on we can try to reconstruct the dat files from source and then the completion flag will prompt a checksum between the built dat and the original.