AI Protocol Parser: add ParserConfig for the wuffs JSON parser

[tyxia]

Introduces ParserConfig to bound the parsing and capture and control streaming JSON field
extraction on top of WuffsJsonCursor.
The use of ParserConfig are shown in the test handler, agentic(MCP/A2A) and LLM(openAI/Gemini) handler will be implemented in the next PRs.

Capture limitation, (TODO define good defaults)

• max_body_bytes: pre-feed cap on raw bytes entering the cursor.
• max_per_scalar_bytes: per-string-value cap. Over-budget values are
  rejected — dropped entirely
• max_total_scalar_capture_bytes: body-wide retention budget across all captured
  values. A value that would overflow the total is dropped without consuming
  budget, so later smaller values still fit.
• max_element_capture_bytes: per container byte-range capture.

Two capture modes

1. Spec-based extraction (extract_fields) — the operator declares exactly
   which fields to extract as pattern paths ("model", "params._meta.traceparent",
   "messages[].role", "tools[].function.name").

2. Capture-all mode (capture_all_scalars) — captures every scalar value
   (strings, numbers, booleans, nulls) at every depth to maxDepth

Clean-up and Clarification

• clean up and polish the code comment
• buildIndex/PatternPath and path tracking are not needed anymore, replaced by matchesPatternPath
• number token is also bounded by max_body_bytes introduced in this PR.

[repokitteh-read-only]

As a reminder, PRs marked as draft will not be automatically assigned reviewers,
or be handled by maintainer-oncall triage.

Please mark your PR as ready when you want it to be reviewed!

🐱

Caused by: #46300 was opened by tyxia.

see: more, trace.

[tyxia]

/retest

[tyxia]

/retest