V2: remove retry policy from the LLM package #34992

Context

@opencode-ai/llm should execute one provider-turn attempt per generate or stream call. Retry policy belongs to the consumer, matching packages/opencode, where the integration catches failures and decides whether to retry.

Scope

• Remove automatic retry scheduling from packages/llm/src/route/executor.ts.
• Delete retry delay/count helpers and tests that assert automatic second attempts.
• Keep useful provider failure details for consumers, but do not expose retry hooks or retry lifecycle events.

Acceptance Criteria

• One LLM call sends at most one provider request attempt.
• Retry-related executor tests are removed or rewritten to assert single-attempt failures.
• Errors still preserve enough status/header/body/provider detail for opencode-owned retry policy.