avoid incrementing reference count of function args where possible

[davidhewitt]

This is a micro-optimization to avoid temporary reference count cycles in .call family of functions when passed a Rust tuple containing references such as &Py, &Bound, and Borrowed where the reference is known to be owned by code somewhere higher up the stack.

Benchmarks will show an oversize effect on call_method functions because I changed these to re-use the method as a Python str rather than passing a Rust &str (which adds a conversion cost).

[codspeed-hq]

Merging this PR will improve performance by ×2.4

⚠️ 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

⚡ 5 improved benchmarks
✅ 100 untouched benchmarks
⏩ 1 skipped benchmark¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	call_1	270 µs	198.6 µs	+35.96%
⚡	call_method_1	709.4 µs	300.3 µs	×2.4
⚡	call_method_one_arg	612.7 µs	271.5 µs	×2.3
⚡	call_method	1,268.2 µs	791.7 µs	+60.19%
⚡	call	607.2 µs	534.8 µs	+13.53%

---

_{Comparing davidhewitt:borrowed-func-args (44c62ab) with main (46d9ca3)}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩