Force inline mul_wide() and reduce() to get rid of performance regression

[fjarri]

Noticed while checking performance in PR #164 that scalar mul() became slower. Seems to be caused by the compiler ceasing to inline mul_wide() and reduce() starting from commit 0586693 on.

Adding this makes Scalar4x64::mul() 10% faster, but has no effect on Scalar8x32 performance on my machine; nevertheless, I inlined these functions there as well for the sake of uniformity (may be important for a 32-bit embedded device). These functions are only used once (in mul()), so it won't degrade compilation.

I usually prefer not to add explicit inlining, but 10% is a pretty solid speed-up...

[codecov-commenter]

Codecov Report

Merging #166 into master will decrease coverage by 0.11%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #166      +/-   ##
==========================================
- Coverage   60.25%   60.14%   -0.12%     
==========================================
  Files          25       25              
  Lines        3555     3555              
==========================================
- Hits         2142     2138       -4     
- Misses       1413     1417       +4     

Impacted Files	Coverage Δ
k256/src/arithmetic/scalar/scalar_4x64.rs	0.00% <ø> (ø)
k256/src/arithmetic/scalar/scalar_8x32.rs	92.45% <ø> (-0.87%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 23def1e...be18643. Read the comment docs.