feat(api): solver service Phase 3b — server endpoints, job definitions, and result cache

Add server-side solver service module with REST API endpoints, database
schema, job definitions, and runner result caching.

New files:
- migrations/021_solver_results.sql: solver_results table with upsert constraint
- internal/db/solver_results.go: SolverResultRepository (Upsert, GetByItem, GetByItemRevision)
- internal/api/solver_handlers.go: solver API handlers and maybeCacheSolverResult hook
- jobdefs/assembly-solve.yaml: manual solve job definition
- jobdefs/assembly-validate.yaml: auto-validate on revision creation
- jobdefs/assembly-kinematic.yaml: manual kinematic simulation job

Modified:
- internal/config/config.go: SolverConfig struct with max_context_size_mb, default_timeout
- internal/modules/modules.go, loader.go: register solver module (depends on jobs)
- internal/db/jobs.go: ListSolverJobs helper with definition_name prefix filter
- internal/api/handlers.go: wire SolverResultRepository into Server
- internal/api/routes.go: /api/solver/* routes + /api/items/{partNumber}/solver/results
- internal/api/runner_handlers.go: async result cache hook on job completion

API endpoints:
- POST   /api/solver/jobs          — submit solver job (editor)
- GET    /api/solver/jobs          — list solver jobs with filters
- GET    /api/solver/jobs/{id}     — get solver job status
- POST   /api/solver/jobs/{id}/cancel — cancel solver job (editor)
- GET    /api/solver/solvers       — registry of available solvers
- GET    /api/items/{pn}/solver/results — cached results for item

Also fixes pre-existing test compilation errors (missing workflows param
in NewServer calls across 6 test files).

internal/db/jobs.go

@@ -328,6 +328,55 @@ func (r *JobRepository) ListJobs(ctx context.Context, status, itemID string, lim
 	return scanJobs(rows)
 }
 
+// ListSolverJobs returns solver jobs (definition_name LIKE 'assembly-%') with optional filters.
+func (r *JobRepository) ListSolverJobs(ctx context.Context, status, itemID, operation string, limit, offset int) ([]*Job, error) {
+	query := `
+		SELECT id, job_definition_id, definition_name, status, priority,
+		       item_id, project_id, scope_metadata, runner_id, runner_tags,
+		       created_at, claimed_at, started_at, completed_at,
+		       timeout_seconds, expires_at, progress, progress_message,
+		       result, error_message, retry_count, max_retries,
+		       created_by, cancelled_by
+		FROM jobs WHERE definition_name LIKE 'assembly-%'`
+	args := []any{}
+	argN := 1
+
+	if status != "" {
+		query += fmt.Sprintf(" AND status = $%d", argN)
+		args = append(args, status)
+		argN++
+	}
+	if itemID != "" {
+		query += fmt.Sprintf(" AND item_id = $%d", argN)
+		args = append(args, itemID)
+		argN++
+	}
+	if operation != "" {
+		query += fmt.Sprintf(" AND scope_metadata->>'operation' = $%d", argN)
+		args = append(args, operation)
+		argN++
+	}
+
+	query += " ORDER BY created_at DESC"
+
+	if limit > 0 {
+		query += fmt.Sprintf(" LIMIT $%d", argN)
+		args = append(args, limit)
+		argN++
+	}
+	if offset > 0 {
+		query += fmt.Sprintf(" OFFSET $%d", argN)
+		args = append(args, offset)
+	}
+
+	rows, err := r.db.pool.Query(ctx, query, args...)
+	if err != nil {
+		return nil, fmt.Errorf("querying solver jobs: %w", err)
+	}
+	defer rows.Close()
+	return scanJobs(rows)
+}
+
 // ClaimJob atomically claims the next available job matching the runner's tags.
 // Uses SELECT FOR UPDATE SKIP LOCKED for exactly-once delivery.
 func (r *JobRepository) ClaimJob(ctx context.Context, runnerID string, tags []string) (*Job, error) {