Replace the manual Real AudioV4 field sequence with generated serial tables

[swackhamer]

Real AudioV4 copied a 31-slot field sequence into its parser. This change removes that sequence and reads the source-generated AudioV4 descriptor through the shared serial reader from #757. The carrier retains signature/version selection and the native 512-byte bounded read.

The shared behavior also corrects legacy output: native UTF-8 question-mark repair fixes Copyright on the real sample, strings stop at NUL, and an incomplete Artist span no longer invents later Copyright/Comment fields. Version 3/5 activation, native warnings and the existing group-0/group-2 occurrence limits remain explicit residuals.

Validation completed locally:

• Six carrier tests, all-feature Clippy with warnings denied, and formatting pass.
• Fresh pinned native/control/candidate runs cover 19 fixtures: all 15 scored candidate Real-* projections match native; ten correct legacy behavior, five are unchanged, and four pre-existing scope/diagnostic cases remain recorded. Exit codes are unchanged.
• A copied ExifTool 13.59 change from AudioV4 Title to UpgradeTitle passes fresh dump/generation/independent verification. Compiling the generated artifact with the unchanged carrier changes exactly that actual output key on Real.ra. No tag-specific Python/Rust edit is needed.
• Shared-cache CLI builds take 1.486 seconds for control and 7.443 seconds for candidate; the copied-source runtime proof compiles in 15.461 seconds.

The complete paired 4,238-file census is independently accepted. With unchanged conformance.py scoring and pinned ExifTool 13.59, correct rows rise 468,086 -> 468,087 and VALUE differences fall 421 -> 420; MISSING 12,240 / RENAME 22 / EXTRA 1,560 are unchanged. The only raw or scored per-file change is Real.ra Copyright now matching native question-mark repair; the other 4,237 files are unchanged. All input hashes are stable and there are zero parse/crash failures. Raw stdout/stderr and exits are retained for all three tools on every file.

The supervisor returns 1 because native ExifTool reports File is empty for the pre-existing zero-byte FujiFilmISPro.jpg. That parsed diagnostic row remains in the denominator; both OxiDex builds exit 0 with identical meaningful output. Independent review accepts this documented diagnostic separately from new failures. Full pair elapsed 273.168 seconds. Production source at PR head is identical to the measured candidate; subsequent Rust changes add only regression assertions.

All five required hosted checks passed at 4bcda9d9: 757 canonical Python tests with zero failures/skips in 643.545 seconds; nextest 5,761 passed with 59 intentional skips; complete Cargo test/doc and explicit native replay passed. Benchmarks were intentionally skipped. This PR claims no new project autogenerated percentage; count the manual reader and 31-slot sequence as retired only after acceptance and landing.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.