Qualify 1.2.1 test coverage and schema failure handling

[ThorFuchs]

This proposes the 1.2.1 qualification patch against release/1.2, rooted at published 1.2.0 (25fb04f). Main advanced to JSONHTML/1.3 in PR #21; the maintainer explicitly approved this maintenance destination so that later features and sibling work remain untouched. Best-effort schema diagnostics now preserve ordinary invalidity when optional output exceeds its byte cap; required work/annotation failures remain terminal. A completely missing native schema bundle now throws structured metaSchema failure instead of trapping, with all nine integrity pins retained.

Add 126 meaningful Swift Testing functions (318 roots / 173 parameterized invocations), including foreign/native Codable, exact values and failures, causal streaming/backpressure, schema/regex/Patch decisions and 20 public end-to-end functions. Coverage exports preserve five module denominators and exact six-target discovery, reject missing objects/saturated counters, and run on both Linux architectures. Fixed-seed replays and a 26-mutant native pilot supplement behavior assertions. No public API/product, dependency, profile, platform or 1.3.0 change.

Validation already materialized against matching implementation/test inputs:

• Native debug/release/full ASan: 318 functions / 173 parameterized invocations; targeted TSan: 32 functions; strict-memory safety, 49 cap probes and public consumers passed.
• All five native modules meet the 98% authored-line / 95% region review targets without denominator exclusions. ARM64/x86_64 Linux qualification, safety and complete coverage passed.
• WASI: Node full debug/release roots; Node and WasmKit affected QA (145 functions per configuration); strict memory compilation. No new WASI sanitizer/corpus claim.
• Independent suites: 287/318 JSON syntax cases evaluated successfully plus 31 policy cases; 2300/2337 schema cases evaluated successfully plus 37 policy cases. Policy cases are not evaluated successes.
• Fresh closeout: 12 harness tests, public/privacy and whitespace checks passed. Fresh native DocC lane passed: 582/582 authored public symbols and named parameters, empty diagnostics, 10 executed README examples. This PR reruns current Native and Linux CI.

Performance remains explicitly qualified by host conditions. The maintainer requested a matched run despite competing CPU load: ten processes / five alternating pairs measured 0.5761326 ms parent vs 0.5664125 ms current (−1.6871%); no sample discarded and work/logical-retention counters matched. Fifty observations recorded 2.61–58.56% idle CPU and load1 21.98–31.69 on ten cores. This does not resolve the earlier +11.9833% comparison or establish an intrinsic improvement. The unchanged 10% review trigger and outstanding quiet-host qualification remain visible; historical streaming disposition is unchanged.

See review readiness and receipts and acceptance outcomes. The maintainer authorized merging after source review and all Native/Linux checks pass. Quiet-host performance qualification remains a separate release follow-up; the loaded-host observation is not relabeled as controlled evidence. Existing physical-iPad deferral and WASI limitations remain. Merging, tagging and publication are separate actions.