Refactor fe_isnonzero to fe_isnonzero_vartime

[tisonkun]

This refers to #1827.

An intuitive solution according to @briansmith's comment. But I know little about crypto so perhaps some invariants are broken.

[briansmith]

Thanks for tackling this.

[codecov]

Codecov Report

Attention: Patch coverage is 88.88889% with 1 line in your changes missing coverage. Please review.

Project coverage is 96.30%. Comparing base (9cb93e0) to head (407599a).
Report is 174 commits behind head on main.

Files with missing lines	Patch %	Lines
crypto/curve25519/curve25519.c	88.88%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1893   +/-   ##
=======================================
  Coverage   96.29%   96.30%           
=======================================
  Files         135      135           
  Lines       20663    20667    +4     
  Branches      226      228    +2     
=======================================
+ Hits        19898    19903    +5     
  Misses        730      730           
+ Partials       35       34    -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[tisonkun]

Thanks for your review @briansmith!

Do you think it's better to exclude changes around src/constant_time.rs in this PR and move forward, or we can wait for #1899 merged?

[briansmith]

Yes, I think we should limit this to the curve25519.c change and mem.h deletion.

[tisonkun]

Updated. PTAL @briansmith.

[tisonkun]

@briansmith ping as a reminder :D

[briansmith]

I noticed that the code coverage for has missing coverage for one of the calls. I am investigating that.

[tisonkun]

@briansmith ping as a reminder :D

[tisonkun]

@briansmith I suppose the codecov regression is false positive..

[tisonkun]

@briansmith is there other blocker to this PR?