Fix regression in float logic

[leonardt]

In PR #81 the magma definition of the reinterpret functions did not perform the required type conversions, which caused the RTL to generate standard add/mul instead of float add/mul

[leonardt]

@cdonovick @rdaly525 there seems to be a regression introduced by the new FP functional model. Reverting the functional model via this commit 2651391 fixes the RTL test for me, so that would suggest that the new FP functional model is exhibiting the incorrect behavior.

[leonardt]

Seems to be fixed by merging the changes in #92

[leonardt]

@cdonovick now the fp tests are failing only when using the assembler (works when using the low level tuple interface), any hunches as to why this might be occurring? I'm going to poke around and see what the encoding values are

[leonardt]

I think I found the issue from the ncsim output:

inst = 21040819108348690454;
--
  | \|
  | ncvlog: *W,INTOVF (WrappedPE_tb.sv,37\|34): bit overflow during conversion from text [2.5(IEEE)] (32 bits).

need to format large integer literal in a way that ncsim accepts

[leonardt]

Issue fixed by fault generating the appropriate integer literal specifiers for width in this branch: leonardt/fault#107