Floating-point changes.

`fpX_to_value` is defined as `float_to_value o fpX_to_float` (previously this was called `fpX_to_real`, which was misleading)
`fpX_to_real` is now defined as `float_to_real o fpX_to_float`.

This change will break proofs that use the old names.

Check for signalling NaNs in floating-point conversions.

Add some more float-to-float functions to machine_ieeeTheory.

Also add versions that output flags for the exception status.

Avoid potential UNCHANGED exception when evaluating 64-bit float operations.

Problem identified by Yong Kiam.

Update floating-point theory to specify exception flags for some operations.

• The flags are: DivideByZero, InvalidOp, Overflow, Precision and Underflow.
• The arithmetic operations now return a pair (flags and result).
• Native (oracle) evaluation has been overhauled. It now only applies to some 64-bit machine_ieeeTheory operations directly.
• The L3 Importer has been updated.

Improve handling of NaNs in binary_ieeeTheory.

The choice of NaN is now a function of the operation being performed. Previously all instances of float_some_nan would be the same (for any given float type).

Revert "holyhammer: fix escaping of 0 and 1"

This partially reverts commit 996d244090d790208dd6d865548a40acfb0e1e04.
It seems like there were mistaken changes to an irrelevant file in that
commit.

holyhammer: fix escaping of 0 and 1

Update for latest version of L3.

Now supports floating-point square root and size conversions.

Relocate examples/l3-machine-code/common/fp to src/floating-point.