Add ApiConfig::validateJson to skip WebsocketMessageParser's DOM validation pass

[TuxedoFish]

Summary

• WebsocketMessageParser::crack() runs a full simdjson DOM parse ahead of its on-demand parse, solely to guarantee malformed input is rejected as a catchable error rather than hitting an internal simdjson ondemand assertion (Malformed JSON in parseAllPerpMetas can abort the process via a simdjson internal assertion #109).
• Adds ApiConfig::validateJson (default true, non-breaking) to skip that extra pass once the feed is trusted.

Test plan

• ctest (29/29 passing, including websocket_parser_test)
• Latency comparison against hyperliquid-md-benchmark (results in PR comment)

🤖 Generated with Claude Code

https://claude.ai/code/session_018Le73N2EZjGLsCqmCScAse

[TuxedoFish]

Latency impact

Measured with hyperliquid-md-benchmark — its replay harness times WebsocketMessageParser::crack() end-to-end on real captured mainnet frames. Pointed its FetchContent at this branch, added a [validateJson=1] CLI arg to bench_cpp that sets ApiConfig::validateJson before constructing the parser, then ran both channels 3x each with validateJson on vs off.

channel	validateJson	n	mean (ns)	p50	p90	p99	p99.9
l2Book	1 (default)	12,600	6535	5569	9089	9884	16315
l2Book	0	12,600	6476	5570	9112	10003	16541
bbo	1 (default)	20,000 (x3)	844–858	779–822	853–1039	1320–1654	3317–4899
bbo	0	20,000 (x3)	475–549	460–471	488–888	745–1037	1127–1594

• l2Book: no real difference — those messages hit crackL2BookFast and return before validateStructure is ever reached, so this flag doesn't touch that path.
• bbo (and every other channel without a fast path — trades, orderUpdates, userFills, allMids, etc.): consistent ~35-45% reduction in mean/p50/p90 across all three repeated runs.

levelsSeen output matched exactly between the two settings on the same captured data, confirming no functional regression — this is purely the cost of the extra DOM pass.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 70.29%. Comparing base (cc645fd) to head (e3e82a1).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #158   +/-   ##
=======================================
  Coverage   70.28%   70.29%           
=======================================
  Files          54       54           
  Lines        8898     8900    +2     
  Branches      539      539           
=======================================
+ Hits         6254     6256    +2     
  Misses       2639     2639           
  Partials        5        5           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.