[code-reflection] RFR: [hat][bfloat16] round-to-nearest-even implemented

[Juan Fumero]

Extension of the `bfloat16` type to support round-nearest-to-even. This improves precision for some of the test cases. 

How to test? 


 HAT=SHOW_CODE java -cp hat/job.jar hat.java test ffi-cuda hat.test.TestBFloat16Type

 HAT=SHOW_CODE java -cp hat/job.jar hat.java test ffi-opencl hat.test.TestBFloat16Type


This patch has been checked for MacOS (opencl-only) and Linux (OpenCL and CUDA).

Thank @jddarcy for the advise on this.

-------------

Commit messages:
 - [hat][bfloat16] round-to-nearest-even implemented

Changes: https://git.openjdk.org/babylon/pull/830/files
  Webrev: https://webrevs.openjdk.org/?repo=babylon&pr=830&range=00
  Stats: 116 lines in 7 files changed: 63 ins; 19 del; 34 mod
  Patch: https://git.openjdk.org/babylon/pull/830.diff
  Fetch: git fetch https://git.openjdk.org/babylon.git pull/830/head:pull/830

PR: https://git.openjdk.org/babylon/pull/830