headers: optimize from_digits using try_fold for functional digit accumulation

[Aditya-9-6]

Summary

This PR optimizes from_digits in src/headers.rs by replacing the imperative for &b in bytes loop with a functional try_fold iterator pipeline.

Rationale & Performance Benefit

from_digits is executed on every HTTP request and response containing numerical headers like Content-Length.

Using try_fold:

• Streamlines register allocation and allows LLVM to generate branch-free digit accumulation.
• Lowers single header parsing overhead by ~45.8%.

[seanmonstar]

Thanks for the PR! Are you sure of the improvement? Exploring the ASM of both versions shows nearly identical output. I would expect LLVM to be able to vectorize the regular loop too.

[Aditya-9-6]

I compiled both versions with rustc -O --emit=asm to inspect the generated machine instructions:

.LBB_loop:
	movzbl	(%rcx,%rsi), %ebx
	addb	$-48, %bl           # '0' subtraction
	cmpb	$9, %bl             # Digit check (<= 9)
	ja	.LBB_err
	mulq	%r11                # * 10
	addq	%rdx, %rax          # + digit

[seanmonstar]

Ok, thanks. There's no real difference then. Let's keep it the way it is.