fix(repository): replace peek_result panic with InfrastructureError

[SAY-5]

Summary

Closes #307.

peek_result (api/src/apps/repository.rs) panics via unreachable!() when the database returns a row whose status = 'done' but whose result_success and result_error columns (and the merged-with task's columns) are all NULL. The incident in #307 shows this state does occur in production — the operator's database had ten such rows:

SELECT id, created_at FROM app_task a
WHERE result_success is null AND result_error is null AND a.status = 'done';
-- 10 rows

The unreachable!() therefore reduced PREvant to a panicking task fetcher whenever a corrupt row was peeked.

Fix

Replace the unreachable arm with Err(AppsError::InfrastructureError { error: ... }) so the caller can return a 500, the operator can inspect/repair the row, and the rest of the system keeps running:

_ => Err(AppsError::InfrastructureError {
    error: format!(
        "Task {status_id} is marked done but has neither a success nor an error result stored"
    ),
}),

The error variant is already exported and is the closest semantic match for "the task ran but its outcome wasn't persisted".

Test plan

• cargo build -p prevant — passes
• Maintainer verification on the live staging environment (or a manually-corrupted app_task row) that the API now returns 500 with the new error message instead of crashing the worker.