feat(cargo-codspeed): add internal build mode

[not-matthias]

No description provided.

[codspeed-hq]

Merging this PR will degrade performance by 19.27%

⚡ 1 improved benchmark
❌ 16 regressed benchmarks
✅ 558 untouched benchmarks

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

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	Iterative[20]	20 ns	21 ns	-4.76%
❌	WallTime	Iterative	20 ns	21 ns	-4.76%
❌	WallTime	sum_fold	305 ns	360 ns	-15.28%
❌	WallTime	Iterative[21]	21 ns	22 ns	-4.55%
❌	WallTime	iter_batched_per_iteration	36 ns	40 ns	-10%
❌	WallTime	from_elem_decimal[1024]	80 ns	83 ns	-3.61%
⚡	WallTime	string_copy_with_bytes_counter	61 ns	37 ns	+64.86%
❌	WallTime	graph_coloring[5]	1.1 µs	1.2 µs	-9.51%
❌	WallTime	sleep_1ms	1 ms	1.1 ms	-3.3%
❌	WallTime	fib_in_thread_bench_local[30]	6.8 ms	7 ms	-3.12%
❌	WallTime	fib_in_thread[30]	6.8 ms	7 ms	-3.28%
❌	WallTime	fib_in_thread_bench[30]	6.8 ms	7 ms	-3.34%
❌	Simulation	find_highest_set_bit[255]	122.2 ns	151.4 ns	-19.27%
❌	Simulation	find_highest_set_bit[65535]	122.2 ns	151.4 ns	-19.27%
❌	Simulation	find_highest_set_bit[1024]	122.2 ns	151.4 ns	-19.27%
❌	Simulation	find_highest_set_bit[42]	122.2 ns	151.4 ns	-19.27%
❌	Simulation	process_items_with_counter	806.7 ns	835.8 ns	-3.49%

---

_{Comparing cod-1865-improve-dx-of-memory-profiling-alias-with-instrumentation (693edb3) with main (0afd168)}

[GuillaumeLagrange]

olgtm, assuming skipped benchmark are just because we need to rebase