Compiled ReverseDiff for training on CPU

[avik-pal]

Fixes https://github.com/LuxDL/Lux.jl/issues/642

For performance https://github.com/SciML/ADTypes.jl/pull/63 will be needed

• [x] Consolidated caching for parameters
• [x] Caching Parameters
  • [x] Tracker
  • [x] ReverseDiff
• [x] Compiled ReverseDiff
• [x] Type-Piracy of Tracker has been removed. @test_gradients might fail need to verify.

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 93.33333% with 9 lines in your changes missing coverage. Please review.

Project coverage is 97.01%. Comparing base (66bd131) to head (9eb4def).

:exclamation: Current head 9eb4def differs from pull request most recent head 1da2e52

Please upload reports for the commit 1da2e52 to get more accurate results.

Files	Patch %	Lines
ext/LuxReverseDiffExt/training.jl	93.75%	4 Missing :warning:
src/contrib/training.jl	20.00%	4 Missing :warning:
src/utils.jl	92.85%	1 Missing :warning:

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #722 +/- ## ========================================== - Coverage 97.32% 97.01% -0.32% ========================================== Files 52 53 +1 Lines 2653 2711 +58 ========================================== + Hits 2582 2630 +48 - Misses 71 81 +10 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.