Add unicodedata.category benchmark script

Add a small Python-level benchmark under Tools/unicode for comparing
unicodedata.category() lookup speed across builds on three fixed
workloads: all code points, BMP only, and ASCII only.

Current results from optimized non-debug builds (-O3 -DNDEBUG),
comparing clean HEAD vs the packtab branch:
- all: baseline 98.98 ns median, packtab 108.44 ns median
- bmp: baseline 97.44 ns median, packtab 105.01 ns median
- ascii: baseline 83.80 ns median, packtab 82.53 ns median

Author: behdad

Tools/unicode/benchmark_unicodedata_category.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+"""Benchmark Python-level unicodedata.category() lookups.
+
+Runs three fixed workloads:
+- all Unicode code points
+- BMP only
+- ASCII only
+"""
+
+from __future__ import annotations
+
+import statistics
+import time
+import unicodedata
+
+
+LOOPS = 5
+SAMPLES = 7
+DATASETS = {
+    "all": "".join(map(chr, range(0x110000))),
+    "bmp": "".join(map(chr, range(0x10000))),
+    "ascii": "".join(map(chr, range(0x80))),
+}
+
+
+def run_once(chars: str) -> tuple[float, int]:
+    category = unicodedata.category
+    checksum = 0
+    t0 = time.perf_counter()
+    for _ in range(LOOPS):
+        for ch in chars:
+            gc = category(ch)
+            checksum += ord(gc[0]) + ord(gc[1])
+    elapsed = time.perf_counter() - t0
+    return elapsed, checksum
+
+
+def benchmark(name: str, chars: str) -> None:
+    lookups = len(chars) * LOOPS
+
+    # Warm up specialization and caches before timing.
+    run_once(chars)
+
+    samples = []
+    checksum = None
+    for _ in range(SAMPLES):
+        elapsed, checksum = run_once(chars)
+        samples.append(elapsed)
+
+    best = min(samples)
+    median = statistics.median(samples)
+    mean = statistics.fmean(samples)
+
+    print(f"dataset: {name}")
+    print(f"codepoints: {len(chars)}")
+    print(f"lookups/sample: {lookups}")
+    print(f"checksum: {checksum}")
+    print(f"best_s: {best:.6f}")
+    print(f"median_s: {median:.6f}")
+    print(f"mean_s: {mean:.6f}")
+    print(f"best_ns_per_lookup: {best * 1e9 / lookups:.2f}")
+    print(f"median_ns_per_lookup: {median * 1e9 / lookups:.2f}")
+    print()
+
+
+def main() -> None:
+    print(f"python: {unicodedata.unidata_version=}")
+    print(f"samples: {SAMPLES}")
+    print(f"loops: {LOOPS}")
+    print()
+
+    benchmark("all", DATASETS["all"])
+    benchmark("bmp", DATASETS["bmp"])
+    benchmark("ascii", DATASETS["ascii"])
+
+
+if __name__ == "__main__":
+    main()