perf(benchmarks): add ttl_cache microbench (#238)

Summary
- Add a ttl_cache microbenchmark to track cached call overhead with a
frozen timer.

Rationale
- Freezing the cache timer keeps the benchmark in the cache-hit path
even on slower CI, yielding stable results.

Details
- New `benchmarks/test_cache_benchmarks.py` adds a ttl_cache microbench
using a frozen timer function.
- Tests: `python setup.py build_ext --inplace` (mypy-compiled
extensions; verified `evmspec/_new.cpython-314-x86_64-linux-gnu.so`),
`python -m pytest benchmarks/`.

Author: BobTheBuidler

benchmarks/test_cache_benchmarks.py

@@ -0,0 +1,19 @@
+# mypy: disable-error-code=misc
+import pytest
+from pytest_codspeed import BenchmarkFixture
+
+from benchmarks.batch import batch
+from evmspec.data._cache import ttl_cache
+
+
+def _frozen_time() -> float:
+    return 0.0
+
+
+@pytest.mark.benchmark(group="ttl_cache_call")
+def test_ttl_cache_call(benchmark: BenchmarkFixture) -> None:
+    @ttl_cache(maxsize=32, ttl=1, timer=_frozen_time)
+    def _cached_call(x: int) -> int:
+        return x + 1
+
+    benchmark(batch, 25_000, _cached_call, 1)