Fix repair retry accounting and replay recovery

[NeerajGodiyal]

Summary

Fix several failure paths in repair, replay shutdown and saved-state recovery:

• Correct repair reservation accounting across retries, cancellation and failed writes, and prevent local attempt IDs from wrapping after 256 attempts.
• Return replay failures through the CLI. A fetch stall returns exit code 75 only after successful cleanup; persistence or cleanup failures prevent retry eligibility.
• Wait for receiver and background work before closing their resources, propagate cleanup errors, and stop replay workers and queued rate-limit waits safely.
• Accept valid saved progress before the first durable fold without a false corruption error, and retry state persistence after a failed post-fold save.

This does not change restart policies or include the separate logger lifecycle changes in #267.

Testing

• All ten affected packages, repeated regression tests, race checks and full-tree go vet passed.
• Native Linux A/B tests used the actual CLI with copied snapshot/accounts data:
  • Fetch stall: upstream exited 0; the fix exited 75 after cleanup.
  • State-save failure: upstream exited 0; the fix exited 1.
  • Valid saved-state reopening: upstream rejected it; the fix resumed.
  • Graceful shutdown completed with exit 0.
• All four stall/save-failure runs executed the same 3,023-transaction block and produced the same bank hash.

The full repository suite still has three failures reproduced on unchanged upstream: missing conformance vectors, a genesis archive decoding error, and a network-dependent RPC test under network isolation.

These checks cover the failure and recovery paths above, not a long-running voting soak.