fix: honor ls_provider for gen_ai.system in OTel translator [closes #2936]

[langsmith-fleet]

Description

wrapAnthropic explicitly emits ls_provider: "anthropic" via getInvocationParams, but LangSmithToOTELTranslator.setGenAiSystem was ignoring it and only inferring gen_ai.system from model-name heuristics. For model aliases like "claude-haiku-4-5" the heuristic works (starts with "claude"), but this is fragile — any alias that doesn't happen to match a pattern causes gen_ai.system to fall back to "langchain", and Playground cannot resolve the Anthropic model dropdown.

The fix: check ls_provider in runInfo.extra.metadata first (authoritative, set explicitly by all wrappers), then fall back to model-name heuristics only when it's absent. A static LS_PROVIDER_TO_GEN_AI_SYSTEM map covers all existing providers (anthropic, openai, azure, bedrock, cohere, deepseek, gemini, groq, mistral, perplexity, vertex, xai).

This also benefits wrapOpenAI and any future wrappers for free.

Test Plan

• Unit tests added in js/src/tests/otel_translator.test.ts covering: ls_provider=anthropic, ls_provider=openai, model-name fallback, unrecognized model fallback, and alias-without-"claude" case (the key regression scenario from wrapAnthropic + LANGSMITH_TRACING_MODE=otel traces don't auto-resolve in Playground #2936)

[github-actions]

JS perf benchmark

Lower is better. Noisy on shared runners — treat as a signal, not a gate.

Base64-heavy payload

Single large base64 string per message — the shape the worker-offload path is optimized for.
Payload: 2511.2 KB in / 5.2 KB out, 100 runs.

metric	main	this PR	delta
Wall time (ms)	1929.08	1916.41	-0.7%
createRun total (ms)	73.62	84.04	+14.1%
createRun p50 (ms)	0.63	0.70	+12.4%
createRun p95 (ms)	0.94	0.89	-5.5%
createRun p99 (ms)	19.87	27.19	+36.9%
createRun max (ms)	19.87	27.19	+36.9%
updateRun total (ms)	41.32	41.56	+0.6%
updateRun p95 (ms)	0.68	0.74	+8.9%
loop lag total (ms)	971.48	968.86	-0.3%
loop lag p50 (ms)	0.09	0.10	+0.3%
loop lag p95 (ms)	3.71	3.83	+3.1%
loop lag p99 (ms)	15.32	22.57	+47.4%
loop lag max (ms)	259.18	89.56	-65.4%

Structural payload

Many small strings across a wide/nested object graph. Should bypass worker offload and use sync flush.
Payload: 1239.5 KB in / 13.3 KB out, 100 runs.

metric	main	this PR	delta
Wall time (ms)	1477.49	1472.46	-0.3%
createRun total (ms)	400.76	390.16	-2.6%
createRun p50 (ms)	3.54	3.42	-3.5%
createRun p95 (ms)	6.13	5.84	-4.7%
createRun p99 (ms)	8.82	20.23	+129.5%
createRun max (ms)	8.82	20.23	+129.5%
updateRun total (ms)	33.23	35.06	+5.5%
updateRun p95 (ms)	0.57	0.54	-5.3%
loop lag total (ms)	1026.13	1030.90	+0.5%
loop lag p50 (ms)	0.07	0.08	+12.4%
loop lag p95 (ms)	5.25	4.05	-23.0%
loop lag p99 (ms)	127.64	125.99	-1.3%
loop lag max (ms)	161.76	136.09	-15.9%