fix(reborn): harden capability approval lifecycle

[serrrfirat]

Summary

Follow-up to #3071 after the capability-host base PR landed. Hardens the approval/run-state/lease lifecycle around the review findings:

• Persist spawn approval requests instead of failing terminally when authorize_spawn_with_trust returns RequireApproval.
• Add CapabilityHost::resume_spawn_json so approved spawn requests can claim the matching lease, start the process, consume the lease, and complete run state.
• Add spawn-specific invocation fingerprints so dispatch and spawn approvals cannot share the same replay fingerprint.
• Validate approval correlation_id in addition to action/requester/fingerprint before persisting or resuming approvals.
• Revoke a claimed approval lease when resumed dispatch fails after the claim but before successful runtime dispatch.
• Make filesystem run/approval listings tolerate records deleted after directory listing, preventing rollback/listing races from failing the whole list.

Notes

• Still no concrete runtime/dispatcher/WASM dependency from ironclaw_capabilities.
• Existing dispatch approval path is preserved; approve_dispatch now delegates through a shared action-specific helper, and approve_spawn handles Action::SpawnCapability.
• I self-reviewed this diff directly; no review subagent was used.
• skip-regression-check label and [skip-regression-check] commit marker are applied because the workflow false-negatives on crate-local Rust test changes, even though this PR adds multiple regression tests.

Verification

cargo test -p ironclaw_host_api
cargo test -p ironclaw_approvals
cargo test -p ironclaw_run_state
cargo test -p ironclaw_capabilities
cargo clippy -p ironclaw_host_api -p ironclaw_approvals -p ironclaw_run_state -p ironclaw_capabilities --all-targets -- -D warnings
cargo test -p ironclaw_architecture
cargo fmt --all -- --check
git diff --check

All passed locally.

[gemini-code-assist]

Code Review

This pull request adds support for asynchronous process spawn approvals, aligning it with the existing dispatch approval flow. Key updates include refactoring the approval resolution logic, modifying the capability host to handle pending spawn requests and run-state transitions, and implementing lease revocation on invocation failure. Additionally, filesystem-based stores now ignore missing files during directory scans. Feedback recommends centralizing the duplicated file-reading logic within the run-state stores into a helper function to improve maintainability and consistency.

[gemini-code-assist]

This logic to gracefully handle NotFound errors is duplicated in FilesystemApprovalRequestStore::records_for_scope (lines 826-830). To improve maintainability and ensure consistent behavior, consider extracting this into a centralized helper function that handles errors gracefully by returning None, as per repository guidelines for resource parsing. Additionally, use the let-else pattern to skip missing files without aborting the scan.

Example helper:

async fn read_file_tolerant<F: RootFilesystem>(
    filesystem: &F,
    path: &VirtualPath,
) -> Result<Option<Vec<u8>>, RunStateError> {
    match filesystem.read_file(path).await {
        Ok(bytes) => Ok(Some(bytes)),
        Err(error) if is_not_found(&error) => Ok(None),
        Err(error) => Err(error.into()),
    }
}

Example usage:

let Some(bytes) = read_file_tolerant(self.filesystem, &entry.path).await? else {
    continue;
};

References

1. Centralize parsing logic for resources that can fail into a single function that handles errors gracefully (e.g., by returning None) to ensure consistent behavior across all call sites.
2. Use let Some(...) = ... else { continue; } instead of the ? operator when scanning resources to skip items that don't match the expected format without prematurely aborting the entire scan.