Make pcc allclose tunable for each test. Add test for MLP training.

[umalesTT]

When tensors are too close, pcc will output NaN values. Therefore, for each test it should be possible to separately tune the threshold of allclose.rtol and allclose.atol below which pcc won't be calculated and therefore test will be able to pass.

This is applied in test for MLP neural network training with MSE loss and SGD optimizer.

[github-actions]

	Tests	Passed ✅	Skipped ⚠️	Failed
TT-XLA Tests	457 ran	390 passed	67 skipped	0 failed

Test	Result
No test annotations available

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 77.03%. Comparing base (1b68a3c) to head (f3236bb).
Report is 1 commits behind head on main.

✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #189   +/-   ##
=======================================
  Coverage   77.03%   77.03%           
=======================================
  Files          21       21           
  Lines        1006     1006           
=======================================
  Hits          775      775           
  Misses        231      231           

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