DNM: Benchmark overhead of setdefault calls in _prepare_headers #8288

Job	Run time
Generate llhttp sources	6s
Linter	1m 11s
Benchmark	4m 16s
Test (3.9, ubuntu, false)	4m 11s
Test (3.9, macos, false)	2m 51s
Test (3.9, windows, false)	4m 18s
Test (3.9, Y, ubuntu, false)	4m 29s
Test (3.10, ubuntu, false)	4m 9s
Test (3.10, macos, false)	3m 0s
Test (3.10, windows, false)	4m 39s
Test (3.10, Y, ubuntu, false)	4m 48s
Test (3.11, ubuntu, false)	4m 2s
Test (3.11, macos, false)	3m 8s
Test (3.11, windows, false)	4m 34s
Test (3.11, Y, ubuntu, false)	4m 2s
Test (3.12, ubuntu, false)	4m 9s
Test (3.12, macos, false)	3m 19s
Test (3.12, windows, false)	5m 0s
Test (3.12, Y, ubuntu, false)	4m 46s
Test (3.13, ubuntu, false)	4m 2s
Test (3.13, macos, false)	2m 54s
Test (3.13, windows, false)	5m 19s
Test (3.13, Y, ubuntu, false)	4m 18s
Test (pypy-3.9, Y, ubuntu, false)	7m 32s
Test (ubuntu, 3.14, true, Y)	2m 33s
check	2s
Pre-Deploy	0s
Build wheels on ${{ matrix.os }} ${{ matrix.qemu }}	0s
Tarball	0s
Deploy	0s
	1h 37m 38s

Provide feedback