Fix exp2f_rcp to properly handle nan and 0xFE cases

[kainzhong]

Description

Fix the implementation of exp2f_rcp

Fixes #2408

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

Please list the changes introduced in this PR:

Our previous implementation of exp2f_rcp is not accurate. After some discussion we decide to

• If biased_exp is nan, return nan
• Otherwise, return the reciprocal using the fast math trick
  • If biased_exp == 254 (2^127), this is the only case where we have to use the mantissa bits to represent its reciprocal 2^-127, so we directly return the hardcoded result here
  • Otherwise we use a bits shifting trick to obtain the float reciprocal result

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes