feat: add benchmark filtering (--max-size, --category, --skip-longctx)

Both run-baseline.sh and run-suite.sh now support:
- --max-size GB: skip models larger than N GB (prevents OOM)
- --category LIST: filter by catalog category (smoke,dense,moe)
- --skip-longctx: skip 32K context tests (saves time + memory)
- --reps N: configure repetition count
- --help: shows usage with examples

Safe pre-optimization run: benchmark baseline --max-size 20 --skip-longctx
Full post-optimization: benchmark baseline (no filters, all models + longctx)

Also: 4 new BATS tests for flag parsing (98 total, all passing)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

scripts/benchmark/run-suite.sh

@@ -11,12 +11,33 @@ MODEL_DIR="$(data_dir models)"
 TAG="run"
 BACKENDS_FILTER=""
 MODELS_FILTER=""
+SKIP_LONGCTX=false
+MAX_SIZE_GB=0
+CATEGORY_FILTER=""
+REPS_STANDARD=5
+REPS_LONGCTX=3
 
 while [[ $# -gt 0 ]]; do
     case "$1" in
-        --tag|-t)      TAG="$2"; shift 2 ;;
-        --backends|-b) BACKENDS_FILTER="$2"; shift 2 ;;
-        --models|-m)   MODELS_FILTER="$2"; shift 2 ;;
+        --tag|-t)        TAG="$2"; shift 2 ;;
+        --backends|-b)   BACKENDS_FILTER="$2"; shift 2 ;;
+        --models|-m)     MODELS_FILTER="$2"; shift 2 ;;
+        --skip-longctx)  SKIP_LONGCTX=true; shift ;;
+        --max-size|-s)   MAX_SIZE_GB="$2"; shift 2 ;;
+        --category|-c)   CATEGORY_FILTER="$2"; shift 2 ;;
+        --reps|-r)       REPS_STANDARD="$2"; shift 2 ;;
+        --help|-h)
+            echo "Usage: run-suite.sh [OPTIONS]"
+            echo ""
+            echo "Options:"
+            echo "  --tag NAME           Tag this run (default: run)"
+            echo "  --backends LIST      Comma-separated backend filter"
+            echo "  --models LIST        Comma-separated model filename filter"
+            echo "  --skip-longctx       Skip long-context (32K) tests"
+            echo "  --max-size GB        Only bench models up to this file size in GB"
+            echo "  --category LIST      Comma-separated: smoke,dense,moe (from models.conf)"
+            echo "  --reps N             Standard test repetitions (default: 5)"
+            exit 0 ;;
         *) log_warn "Unknown argument: $1"; shift ;;
     esac
 done
@@ -25,9 +46,6 @@ TS="$(timestamp)"
 RESULT_DIR="$(data_dir benchmarks)/${TAG}-${TS}"
 mkdir -p "$RESULT_DIR"
 
-REPS_STANDARD=5
-REPS_LONGCTX=3
-
 log_header "Benchmark Suite: $TAG"
 log_info "Results: $RESULT_DIR"
 
@@ -60,26 +78,56 @@ if (( ${#available_backends[@]} == 0 )); then
 fi
 log_info "Backends: ${available_backends[*]}"
 
-# Find models
-mapfile -t MODEL_PATHS < <(
+# Find and filter models
+mapfile -t ALL_MODEL_PATHS < <(
     find -L "$MODEL_DIR" -type f -name '*.gguf' -not -name 'mmproj-*' \
         \( -name '*-00001-of-*.gguf' -o -not -name '*-000*-of-*.gguf' \) \
         | sort
 )
 
-if [[ -n "$MODELS_FILTER" ]]; then
-    filtered=()
-    for p in "${MODEL_PATHS[@]}"; do
-        name="$(basename "$p")"
-        if echo "$MODELS_FILTER" | tr ',' '\n' | grep -qi "$name"; then
-            filtered+=("$p")
+MODEL_PATHS=()
+for p in "${ALL_MODEL_PATHS[@]}"; do
+    local_name="$(basename "$p")"
+
+    # Name filter
+    if [[ -n "$MODELS_FILTER" ]]; then
+        if ! echo "$MODELS_FILTER" | tr ',' '\n' | grep -qi "$local_name"; then
+            continue
         fi
-    done
-    MODEL_PATHS=("${filtered[@]}")
-fi
+    fi
+
+    # Size filter
+    if (( MAX_SIZE_GB > 0 )); then
+        file_size_gb=$(( $(stat -Lc%s "$p" 2>/dev/null || echo 0) / 1073741824 ))
+        if (( file_size_gb >= MAX_SIZE_GB )); then
+            log_info "Skipping $local_name (${file_size_gb} GB > ${MAX_SIZE_GB} GB limit)"
+            continue
+        fi
+    fi
+
+    # Category filter
+    if [[ -n "$CATEGORY_FILTER" ]] && [[ -f "$PROJECT_ROOT/configs/models.conf" ]]; then
+        matched=false
+        found_in_catalog=false
+        while IFS='|' read -r name repo file size_gb category desc; do
+            [[ "$name" =~ ^#.*$ || -z "$name" ]] && continue
+            if [[ "$local_name" == "$file" ]]; then
+                found_in_catalog=true
+                echo "$CATEGORY_FILTER" | tr ',' '\n' | grep -qF "$category" && matched=true
+                break
+            fi
+        done < "$PROJECT_ROOT/configs/models.conf"
+        if $found_in_catalog && ! $matched; then
+            log_info "Skipping $local_name (category not in: $CATEGORY_FILTER)"
+            continue
+        fi
+    fi
+
+    MODEL_PATHS+=("$p")
+done
 
 if (( ${#MODEL_PATHS[@]} == 0 )); then
-    log_error "No models found. Run: make benchmark-setup"
+    log_error "No models matched filters. Run: make benchmark-setup"
     exit 1
 fi
 log_info "Models: ${#MODEL_PATHS[@]}"
@@ -115,6 +163,9 @@ for MODEL_PATH in "${MODEL_PATHS[@]}"; do
         fi
 
         # Long-context test
+        if $SKIP_LONGCTX; then
+            continue
+        fi
         OUT_LC="$RESULT_DIR/${MODEL_NAME}__${BACKEND_SAFE}__fa1__longctx32768.log"
         if [[ ! -s "$OUT_LC" ]]; then
             printf "\n${BOLD}>> [%s] %s — longctx${RESET}\n" "$BACKEND" "$MODEL_NAME"