Report the latest failed suggestion batch on the deck

The deck reported only a pending or running batch, so a Discover page
opened after a batch failed (or was interrupted by a restart) had no job
to follow and showed "All caught up" or the remaining cards with no
error. GET /api/v1/suggestions/deck now also returns failed_batch: the
newest batch when it failed, read in the same snapshot after orphaned
batches are failed, so at most one of active_batch and failed_batch is
set and a later batch supersedes the failure. The page follows the
failed batch's job like the active one, so its failure shows with the
existing Retry, which asks for a new batch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

migrations/queries/suggestions.sql

@@ -11,6 +11,13 @@ WHERE status IN ('pending', 'running')
 ORDER BY id
 LIMIT 1;
 
+-- name: GetLatestSuggestionBatch :one
+-- The newest batch, whatever its status. Returns sql.ErrNoRows when there
+-- is none.
+SELECT * FROM suggestion_batches
+ORDER BY id DESC
+LIMIT 1;
+
 -- name: GetSuggestionBatchJobID :one
 -- The id of the job that runs batch batch_id.
 SELECT id FROM jobs