Re-apply "[DebugInfo] Emit DW_OP_implicit_value for Floating point constants"

This patch was reverted in 7c182663 due to some failures
observed on PCC based machines. Failures were due to Endianness issue and
long double representation issues.

Patch is revised to address Endianness issue. Furthermore, support
for emission of `DW_OP_implicit_value` for `long double` has been removed
(since it was unclean at the moment). Planning to handle this in
a clean way soon!

For more context, please refer to following review link.

Reviewed By: aprantl

Differential Revision: https://reviews.llvm.org/D83560