Fix json_pipeline_small parse threshold

[TheHypnoo]

Summary

Fixes #437 by raising the lazy JSON parse auto-mode lower bound from 1 KB to 64 KB. The 21 KB json_pipeline_small fixture now stays on the direct parser, avoiding lazy-header and forced-materialization overhead on an iterate-all workload.

This also updates the lazy JSON predicate regression fixture so it still exceeds the auto-mode threshold and refreshes the related docs.

Impact

• json_pipeline_small now runs through the faster direct parser in auto mode.
• The lazy path remains available for targeted tests via PERRY_JSON_TAPE=1.
• Larger auto-mode lazy payloads are still limited to the 64 KB..16 MB window.

Measured Improvement

On the local 20-run issue reproduction:

• Before: issue reported ~47 ms median vs a <=41 ms target.
• After: 28.369 ms median, 1.172 ms standard deviation.
• Output hash remains 7fc66fa8 and output_match=true for all measured runs.

Conflict Resolution

Rebased onto the latest main (0.5.867) and resolved the conflict in Cargo.lock by keeping the current base lockfile. The PR now only changes the runtime threshold, lazy JSON test fixture, and related docs.

Validation

• cargo check --release -p perry-runtime --locked
• Compiled and ran test-files/test_json_lazy_predicates.ts with Perry
• Rebuilt the json_pipeline_small Perry workload and ran 20 measured iterations with the honest_bench output checker