Fix the bug of block attention residuals in AMP

[pkuzyc]

Fix the bug of block attention residuals in AMP. When embedding output is float32 and pushed into the block list, block attention residuals will get a float32-type output, causing a type mismatch error in LinearWithGradAccumulationAndAsyncCommunication backward function.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 0% with 4 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@d479aee). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
src/paddlefleet/transformer/block_attn_res.py	0.00%	1 Missing and 1 partial ⚠️
src/paddlefleet/transformer/transformer_layer.py	0.00%	1 Missing and 1 partial ⚠️

❌ Your patch status has failed because the patch coverage (0.00%) is below the target coverage (90.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff            @@
##             develop    #719   +/-   ##
=========================================
  Coverage           ?   0.00%           
=========================================
  Files              ?       2           
  Lines              ?       4           
  Branches           ?       2           
=========================================
  Hits               ?       0           
  Misses             ?       2           
  Partials           ?       2           

Flag	Coverage Δ
coverage_combine	0.00% <0.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/paddlefleet/transformer/block_attn_res.py	0.00% <0.00%> (ø)
src/paddlefleet/transformer/transformer_layer.py	0.00% <0.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.