Optimize DAG-CBOR encode/decode and CAR block parsing

[MarshalX]

• Drop BufWriter<Vec<u8>>. Write directly into Vec<u8> (VecWriter)
• Pre-filter CID candidates by leading byte before Cid::try_from (looks_like_cid)
• Single dict walk in map encode. No more keys() + values() + by-index lookup
• Exact-type pointer dispatch (Py_TYPE == &PyXxx_Type) instead of is_instance_of MRO walks
• _PyDict_NewPresized(len) for map decode (CPython-only via cfg(CPython) + build.rs; PyPy falls back to PyDict_New); skipped when len ≤ 8
• Drop Cursor wrapper for Cid::read_bytes in decode_car, use &[u8] directly

[codspeed-hq]

Merging this PR will improve performance by 17.98%

⚡ 31 improved benchmarks
❌ 1 regressed benchmark
✅ 160 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	test_dag_cbor_decode_real_data[twitter.json]	6.5 ms	8 ms	-18.41%
⚡	Simulation	test_dag_cbor_encode[roundtrip01.json]	14 µs	12.2 µs	+14.81%
⚡	Simulation	test_dag_cbor_encode[roundtrip02.json]	16 µs	14 µs	+13.98%
⚡	Simulation	test_dag_cbor_encode[roundtrip03.json]	14.2 µs	12.1 µs	+16.61%
⚡	Simulation	test_dag_cbor_encode[roundtrip04.json]	15 µs	13.1 µs	+14.46%
⚡	Simulation	test_dag_cbor_encode[roundtrip05.json]	14.9 µs	12.2 µs	+22.2%
⚡	Simulation	test_dag_cbor_encode[roundtrip06.json]	13.2 µs	11.3 µs	+17.36%
⚡	Simulation	test_dag_cbor_encode[roundtrip07.json]	15 µs	12 µs	+24.71%
⚡	Simulation	test_dag_cbor_encode[roundtrip08.json]	15.1 µs	13.2 µs	+14.57%
⚡	Simulation	test_dag_cbor_encode[roundtrip09.json]	19.1 µs	13.5 µs	+41.59%
⚡	Simulation	test_dag_cbor_encode[roundtrip10.json]	19.3 µs	13.8 µs	+39.69%
⚡	Simulation	test_dag_cbor_encode[roundtrip11.json]	14.9 µs	13.1 µs	+13.54%
⚡	Simulation	test_dag_cbor_encode[roundtrip12.json]	15.2 µs	13.4 µs	+13.95%
⚡	Simulation	test_dag_cbor_encode[roundtrip13.json]	15.1 µs	13.6 µs	+11.21%
⚡	Simulation	test_dag_cbor_encode[roundtrip14.json]	15.7 µs	14 µs	+12.2%
⚡	Simulation	test_dag_cbor_encode[roundtrip15.json]	14.8 µs	13.1 µs	+13.38%
⚡	Simulation	test_dag_cbor_encode[roundtrip16.json]	18 µs	16.3 µs	+10.33%
⚡	Simulation	test_dag_cbor_encode[roundtrip17.json]	15.7 µs	13.9 µs	+13.22%
⚡	Simulation	test_dag_cbor_encode[roundtrip18.json]	15.1 µs	13.5 µs	+12.4%
⚡	Simulation	test_dag_cbor_encode[roundtrip19.json]	15.1 µs	13.4 µs	+12.64%
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing optimize-encode-decode (341f165) with main (2163f07)}

[MarshalX]

Twitter decode is actually faster locally 6.4%. Probably the limitation of codspeed simulation