Unify timeout env vars: OPENAI_TIMEOUT_MS + AGENTMEMORY_LLM_TIMEOUT_MS

[rohitg00]

Two timeout env vars now active across the provider surface after v0.9.17 + #379:

• OPENAI_TIMEOUT_MS — scoped to src/providers/openai.ts (the OpenAI LLM provider from feat: add OpenAI-compatible LLM provider #307). Inline AbortController, default 60s.
• AGENTMEMORY_LLM_TIMEOUT_MS — scoped to the shared src/providers/_fetch.ts helper introduced in fix: add AbortController timeout to all raw-fetch providers #379, used by minimax/openrouter LLM + cohere/gemini/openai/voyage/openrouter embedding. Default 60s.

Same value, different names. Confusing for ops.

Proposal

1. Make AGENTMEMORY_LLM_TIMEOUT_MS the canonical name.
2. OpenAIProvider.call() reads OPENAI_TIMEOUT_MS first (back-compat), then falls back to AGENTMEMORY_LLM_TIMEOUT_MS, then 60s.
3. README documents both in the OpenAI section; quick-start / .env.example shows only AGENTMEMORY_LLM_TIMEOUT_MS.
4. Long-term: per-provider override pattern AGENTMEMORY_<PROVIDER>_TIMEOUT_MS if users ask. Out of scope for now.

Files

• src/providers/openai.ts — line where OPENAI_TIMEOUT_MS is read
• README.md — both env tables
• .env.example — single source of truth

Acceptance

• Setting only AGENTMEMORY_LLM_TIMEOUT_MS=30000 bounds OpenAI provider to 30s.
• Setting both prefers OPENAI_TIMEOUT_MS (back-compat).
• Setting neither uses 60s default.
• New unit test in test/fetch-timeout.test.ts covering the precedence.