fix(engine): harden fallback status ingestion diagnostics and retry classification

[vadimcomanescu]

Summary

Improves status ingestion reliability when canonical stage status is missing or invalid by tightening fallback diagnostics and retry classification.

Problem

• Fallback status ingestion diagnostics were not explicit enough for missing/corrupt/type-mismatched payloads.
• Retry behavior for fallback reads did not clearly separate transient decode corruption from permanent/unreadable conditions.

Why This Change

Failure classification and retry policy must map to real error modes so operators and automation can act deterministically.

What Changed

• Hardened fallback status ingestion path and diagnostics in handlers.
• Added retry policy coverage for transient decode corruption vs permanent read failures.
• Added explicit tests for canonical precedence, fallback validity, missing fallback, and type mismatch paths.

User-Visible Behavior

• Better error diagnostics when status fallback ingestion fails.
• More accurate retry behavior for transient fallback corruption scenarios.
• Deterministic rejection of invalid fallback payload classes.

Risk

• Retry classification changes can alter observed retry count/timing.
• Mitigation: targeted unit coverage for both retryable and non-retryable paths.

Validation

• gofmt and go vet ./...
• go test ./internal/attractor/engine -run 'TestCopyFirstValidFallbackStatus_|TestShouldRetryFallbackRead_|TestCodergenStatusIngestion_' -count=1