Optimize int encode with direct PyLong digit read

[MarshalX]

Adds a CPython 3.12+ fast path in encode_int that reads PyLongObject.lv_tag + ob_digit[0..2] directly to encode 0/1/2-digit ints (≤ 2**60), skipping obj.extract::<i128>() which internally calls _PyLong_AsByteArray with a 16-byte buffer + sign check on every int

Ints requiring 3+ digits, pre-3.12, and non-CPython fall through to the existing i128 path, which still enforces the +-2**64 DAG-CBOR range

[codspeed-hq]

Merging this PR will improve performance by 12.33%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
✅ 191 untouched benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	test_dag_cbor_encode_real_data[citm_catalog.json]	8.1 ms	7.2 ms	+12.33%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing optimize-pylong-int (87d86d2) with main (d2eaf4d)}

[MarshalX]

citm +12% aligns with my local tests; also there are github +4% and twitter +4.6%