perf: use new PyLong* API for num-bigint feature

[chirizxc]

Related: #6040

See codspeed comment for benchmark improvements.

[codspeed-hq]

Merging this PR will improve performance by 68.84%

⚠️ 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

⚡ 23 improved benchmarks
✅ 117 untouched benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	into_bigint_huge_negative	23.6 µs	6.6 µs	×3.6
⚡	extract_biguint_negative_fail	7.2 µs	2.3 µs	×3.2
⚡	extract_biguint_zero	2,758.9 ns	946.1 ns	×2.9
⚡	into_bigint_huge_positive	17.4 µs	6.7 µs	×2.6
⚡	into_biguint_huge	17.1 µs	6.7 µs	×2.6
⚡	into_bigint_big_negative	5.3 µs	2.1 µs	×2.5
⚡	into_bigint_small	3.2 µs	1.5 µs	×2.1
⚡	into_bigint_big_positive	4.5 µs	2.1 µs	×2.1
⚡	into_biguint_big	4.3 µs	2.2 µs	+96.34%
⚡	into_biguint_small	3.3 µs	2 µs	+68.45%
⚡	extract_bigint_small	2.7 µs	1.8 µs	+53.52%
⚡	extract_bigint_huge_negative	13 µs	8.5 µs	+52.37%
⚡	extract_biguint_small	2.6 µs	1.7 µs	+50.52%
⚡	into_biguint_zero	2.9 µs	2 µs	+41.44%
⚡	extract_biguint_huge	11.8 µs	8.5 µs	+37.87%
⚡	extract_bigint_huge_positive	11.7 µs	8.5 µs	+37.36%
⚡	extract_bigint_big_negative	3.9 µs	3.4 µs	+14.46%
⚡	extract_i128_small_pos	1.2 µs	1.1 µs	+14.18%
⚡	extract_i128_zero	1.3 µs	1.1 µs	+13.81%
⚡	extract_i128_small_neg	1.3 µs	1.2 µs	+13.06%
...	...	...	...	...

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

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 chirizxc:PyLongForBigInt (b1899b6) with main (126aa44)}

[davidhewitt]

Thanks, generally looks good on the from-python implementation; I think possible gains yet to be had in the to-python implementation.

[davidhewitt]

Some observations on this path:

• I would prefer to have this logic extracted into a generic function; I think it's easier to maintain than when it's deeply nested inside a macro.
• I think we can avoid the intermediate Vec allocation from the digits and instead wrap the iterator coming out of the big-integer. pylong_from_digits needs an ExactSizeIterator, so it'll be a touch fiddly but I think still possible.
• I think we lack benchmarks for the bigint to-python pathway, so possibly worth landing some of those before we think too hard about this.

[chirizxc]

• I think we lack benchmarks for the bigint to-python pathway, so possibly worth landing some of those before we think too hard about this.

#6148

[davidhewitt]

Also, I think worth giving this a changed newsfragment to note to users we've optimized the bigint conversions on 3.14+.