ChiselGen embeds wrong binary format of float

[kelayamatoz]

In ChiselCodeGen.quoteConst, the string representation of a float is emitted as c.toString. However, when c is of type emul.FloatPoint and when c comes from a LUT, c.toString returns a string of the float format instead of the binary format. As a result, the generated chisel LUT would contain a list of floating point numbers instead of a list of integer numbers that represent the binary form of the floats. This issue mostly affects memories that need to be initialized with high-precision types, e.g. a LUT for doing high-precision sigmoid / tanh functions.

I've put the fix on commit 0ec748d on this branch. It's working with the test apps I pushed, but feel free to let me know if there are other things I'm missing.