Fixes byte order for uint16 and bfloat variables based on endianness

[vibhutisawant]

trafficstars

This PR fixes //tensorflow/python/lib/core:bfloat16_test test case on s390x arch.

Test case initially failed for s390x arch because the little_endian_uint16 and little_endian_bfloat variables were defined with byte order as little-endian. Updating the byte order to Big Endian fixes this test case on s390x. Added an "if guard" to handle the byte order of these variables based on endianness.

[tilakrayal]

@vibhutisawant, Could you please sign CLA. Thank you!

[vibhutisawant]

Hi @tilakrayal I have been covered under corporate CLA(IBM). I see the CLA check in the PR has passed. Please let me know if something is missing. Thanks

[vibhutisawant]

Hi @mattdangerw, could you please take a look? Thanks!

[vibhutisawant]

Hi @wangpengmit could you please have a look at this PR? Thanks!

[gbaned]

Hi @cantonios Can you please review this PR ? Thank you!