[cyrne1-7208]

The 64-bit types section still says double types must not be used, but
-fp64 defaults to on. Update the section to reflect the current
behavior, document -fp64=0, and rename the heading to
64-Bit Floating-Point Types since long/ulong are unaffected by
-fp64.

FP64 support is still incomplete (#321). Document all double
scalar/vector types, including double8/double16 (which require
-long-vector), and note the current limitations of
degrees()/radians() and native math built-ins.

Verified with the local clspv build:

• A double kernel compiles and passes spirv-val by default.
• The same kernel fails to compile with -fp64=0.
• A float kernel still compiles with -fp64=0.
• long/ulong still compile and pass spirv-val with -fp64=0.
• double8/double16 compile and pass spirv-val with -long-vector,
  and fail to compile without it or with -fp64=0.
• degrees()/radians() on double fail spirv-val even with default
  options.
• sin()/exp() on double fail spirv-val with -cl-native-math or
  -use-native-builtins.

[rjodinchr]

Thanks for catching that this section was outdated regarding -fp64! Because FP64 support in clspv is still partial (#321) — e.g. degrees/radians and --cl-native-math still emit invalid SPIR-V for double — I left a couple of suggestions to document FP64 as partially supported across all double scalar/vector types and clarify the section heading.

[cyrne1-7208]

Thanks @rjodinchr for the helpful suggestions and the approval! The general FP64 note is cleaner and less likely to go stale as individual built-ins are fixed, and the heading now correctly scopes the section to floating-point types. I appreciate the review.