fix(ecr,sqs,elasticache): data-durability + portability gaps (bug-hunt 4.2/4.3/4.4)

[vieiralucas]

Summary

Three Tier-2 durability / portability findings from the 2026-07-15 bug hunt, each in a restart or error path CI never exercised.

• 4.2 ECR — orphaned layer uploads 500 on restart. In-progress layer-upload spools live under the system temp dir (spool_path) and never survive a restart, but layer_uploads was persisted, so a restart left orphan entries whose next UploadLayerPart opened a missing spool and returned 500. layer_uploads is now #[serde(skip)] (ephemeral, consistent with the gone spools — a stale client simply re-initiates), and the append opens gained .create(true) as defense-in-depth.
• 4.3 SQS — unresolvable-DLQ redrive hot loop. When a redrive target ARN doesn't resolve (DLQ deleted / typo) the message is correctly returned to the source queue instead of destroyed, but it re-entered with receive_count still past maxReceiveCount and visible_at = None, so the very next ReceiveMessage re-redrove it into the same dead target and spun. The receive history is now reset on return-to-source (mirroring the existing DLQ-landing reset), so the message is reprocessed normally.
• 4.4 ElastiCache — hardcoded /tmp. The snapshot RDB dump path hardcoded /tmp/fakecloud-ec-*.rdb, absent on Windows and sometimes non-writable in containers. Now roots under std::env::temp_dir() via a new snapshot_rdb_temp_path helper.

Test plan

• cargo test -p fakecloud-ecr --lib layer_uploads_are_not_persisted_across_snapshot (new).
• cargo test -p fakecloud-sqs --lib redrive_with_unresolvable_dlq_arn_keeps_message_on_source_queue — extended to assert the returned message is re-receivable (loop-prevention).
• cargo test -p fakecloud-elasticache --lib rdb_temp_path_roots_under_platform_temp_dir (new).
• cargo clippy -p fakecloud-ecr -p fakecloud-sqs -p fakecloud-elasticache --tests -- -D warnings clean.

No API surface / SDK / docs change (persistence + portability internals only).

---

Summary by cubic

Fixes three durability and portability bugs across fakecloud-ecr, fakecloud-sqs, and fakecloud-elasticache. Prevents ECR 500s after restarts, SQS redrive hot loops on missing DLQs, and snapshot path failures on non-Unix systems.

• Bug Fixes
  • ECR: Stop persisting in-progress layer_uploads (#[serde(skip)]) to avoid orphaned state after restart; append opens now use .create(true) to recreate missing spools.
  • SQS: When the DLQ target ARN is unresolvable, reset receive_count and first_received_at before returning the message to the source queue so it is reprocessed, not redriven in a loop.
  • ElastiCache: Write snapshot RDB files under std::env::temp_dir() via snapshot_rdb_temp_path instead of hardcoded /tmp for cross-platform portability.

^{Written for commit 367b47c. Summary will update on new commits.}

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.