Add own implementation of CAR reader

[MarshalX]

• add proper error handling
• add notice about supporting only DAG-CBOR as blocks
• add tests
• investigate performance improvements

[codspeed-hq]

CodSpeed Performance Report

Merging #28 will improve performances by 34.7%

_{Comparing add-car-reader (f0d1449) with main (d8ea506)}

Summary

⚡ 3 improvements
✅ 189 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	add-car-reader	Change
⚡	test_dag_cbor_decode[roundtrip05.json]	35.6 µs	27.9 µs	+27.78%
⚡	test_dag_cbor_decode[roundtrip08.json]	34.3 µs	25.5 µs	+34.7%
⚡	test_decode_car	6.1 s	5.5 s	+10.84%

[MarshalX]

a little bit faster. i expected that because we don't use executor anymore. idk is it possible to do something more (except remove multibasing of CIDs in keys)

[MarshalX]

own impl gains perf boost after upgrading to pyo3 v0.21 (bound api) + pgo