Improve metadata validation and retrieval reliability

## Outcome

Metadata retrieval failures and validation failures are distinguishable, recoverable where appropriate, and understandable to API consumers.

## Scope

- Define typed retrieval, parsing, validation, and persistence errors
- Add bounded retries, timeouts, backoff, and terminal failure behavior
- Preserve diagnostics without exposing unsafe internal details
- Add metrics and failure-mode integration tests

## Acceptance criteria

- [ ] Scope is decomposed into independently deliverable child issues.
- [ ] Architecture and API decisions are linked from this issue.
- [ ] Test, documentation, observability, rollout, and failure behavior are defined where applicable.
- [ ] Epic exit criteria are verified before closure.

## Planning note

This is a 2026 GovTool revamp work item. Its child features and tasks should be added as the scope is refined.

Start by decomposing the retrieval, parsing, validation, persistence, retry, observability, and documentation scope into independently deliverable child issues. Define and link the architecture and API decisions, then specify applicable tests, rollout, and failure behavior. Done means the child work is tracked and the epic exit criteria are verifiable.

Written by the indexing model from the issue text.