fix(strategy): make sentiment telemetry provider-neutral

[Trujillofa]

Summary

Prospective-only telemetry fix: sentiment degradation and event attribution no longer hardcode Grok/xAI when DeepSeek or Z.AI is the configured provider. Successful DeepSeek and Z.AI replies count as live/answered. Failed provider requests are recorded as error_fallback with the configured provider and attempted model. Historical event logs are not rewritten.

Rebased onto current origin/main after #184 (feat/zai-provider) so zai_live is treated as answered and Z.AI failures are not labeled provider=xai.

This PR does not change strategy rules, thresholds, position sizing, trading mode, or execution. Fail-closed BUY blocking and rising-edge-only degradation paging are preserved.

Do not merge as a deploy authorization. Paper evidence window starts only after independent review and a separate deploy.

Type of Change

• Bug fix
• New feature
• Refactoring
• Documentation update
• Tests

SHAs

• Base (origin/main): 7d57c22498aebe75fdf98aa12042176cf07f1273
• Head: dff7e1e7a7ba492438d534312658c5de6b67a21d

What changed and why

File	Why
src/strategy/sentiment_mean_reversion.py	Write error_fallback (provider-neutral); stamp configured provider/model on failures; alert names DeepSeek/xAI/Z.AI, not Grok; shared answered/error source sets (xai_live, deepseek_fallback, zai_live; historical xai_error_fallback).
src/main.py	Honor payload provider on record; infer zai from zai_* sources; daily summary live% uses answered sources.
src/overseer/xai.py	Public provider / model properties so failures can be attributed without poking privates.
scripts/sentiment_report.py	Count successful DeepSeek and Z.AI as live; count both xai_error_fallback and error_fallback as errors; drop Grok-hardcoded health copy.
tests/test_sentiment_mean_reversion.py	DeepSeek 402 attribution, Z.AI error attribution, provider-named alerts, 10-obs 50% gate, fail-closed BUY, pinned trading contract.
tests/test_settings_integration.py	End-to-end DeepSeek 402 payload; Z.AI live provider=zai.
tests/test_sentiment_report.py	Historical parse + DeepSeek/Z.AI success-is-live.

Compatibility for old event logs

Readers accept both old and new source names. Nothing in stored JSONL is rewritten.

Historical / existing source	Parse	Classification
xai_live	yes	answered / live
deepseek_fallback	yes	answered / live (successful DeepSeek; name kept)
zai_live	yes	answered / live (from #184)
xai_error_fallback	yes	no-answer / error
neutral_fallback	yes	no-answer

Prospective writes:

Event	source	provider	model
xAI success	xai_live	xai	answering model
DeepSeek success	deepseek_fallback	deepseek	answering model
Z.AI success	zai_live	zai	answering model
LLM failure	error_fallback	configured provider (deepseek, zai, or xai)	attempted model when the client exposes it, else none
No client	neutral_fallback	none	none

Degradation still uses a rolling 10-observation window at >= 50% no-answer (error_fallback + historical xai_error_fallback + neutral_fallback). Successful DeepSeek and Z.AI replies are not no-answer.

Strategy / trading

Unchanged: RSI/BB gates, sentiment thresholds, panic SELL, high-vol filter, fail-closed block of new BUY when degraded, rising-edge-only Telegram paging. Pinned by test_evaluate_trading_contract_is_unchanged and existing degrade/repage tests.

Testing

• Tests pass locally: uv run pytest -q — 1399 passed
• Lint passes: uv run ruff check .
• Format check: uv run ruff format --check . (ruff is canonical; template black --check not run)
• git diff --check clean

Focused: tests/test_sentiment_mean_reversion.py, tests/test_sentiment_report.py, tests/test_settings_integration.py, tests/test_xai_client.py, tests/test_config.py.

Checklist

• Code follows project style
• Self-review completed
• Tests added/updated
• No merge, deploy, restart, or production mutation from this change
• Documentation updated (if needed) — not needed; telemetry-only

Notes

PR only. After independent review and a separate SHA-pinned deploy, begin a pre-registered DeepSeek paper evidence window. Do not treat the existing three paper trades as validation. Current sentiment-macro config on main is Z.AI; this PR does not change that provider setting.