Fix stack overflow when decoding deeply nested structures like arrays and maps

[MarshalX]

fixes #9

It uses 2 approaches. First for Python 3.8 and second for Python 3.9+. The first one hard limits recursion to 1k (same as Python by default). The second one respects sys.setrecursionlimit. decode_dag_cbor function could now raise RecursionError

CI fails because of temporarily disabled endpoint by bsky team to deal with network load

use of Py_EnterRecursiveCall and Py_LeaveRecursiveCall is slow; so just count depth and check it using Py_GetRecursionLimit; raise PyRecursionError

[codspeed-hq]

CodSpeed Performance Report

Merging #51 will degrade performances by 28.99%

_{Comparing fix-overflow-on-deep-structures (1083c37) with main (7bf9fb9)}

Summary

⚡ 4 improvements
❌ 8 regressions
✅ 180 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	test_dag_cbor_decode[roundtrip01.json]	23.2 µs	16.7 µs	+39.2%
❌	test_dag_cbor_decode[roundtrip22.json]	16.8 µs	22.7 µs	-26%
⚡	test_dag_cbor_encode[roundtrip05.json]	22.9 µs	16.9 µs	+35.38%
❌	test_dag_cbor_encode[roundtrip06.json]	15.1 µs	21.3 µs	-28.99%
⚡	test_dag_cbor_encode[roundtrip11.json]	22.9 µs	17.6 µs	+29.73%
❌	test_dag_cbor_encode[roundtrip12.json]	16.3 µs	18.2 µs	-10.83%
❌	test_dag_cbor_encode[roundtrip16.json]	16.9 µs	22.1 µs	-23.69%
⚡	test_dag_cbor_encode[roundtrip20.json]	22.6 µs	16.6 µs	+35.8%
❌	test_dag_cbor_encode[roundtrip21.json]	16.1 µs	18.2 µs	-11.26%
❌	test_dag_cbor_encode[roundtrip25.json]	16 µs	22 µs	-27.44%
❌	test_dag_cbor_encode[roundtrip26.json]	16.1 µs	18.3 µs	-12.15%
❌	test_dag_cbor_encode_real_data[canada.json]	13.6 ms	15.4 ms	-11.95%

[MarshalX]

R.I.P

[MarshalX]

okay, now it should not affect perf