[Feature Request] Separate compiler flags to disable/enable associative expression rewriting and other FP optimizations

Is your feature request related to a problem? Please describe.
Currently, users have the following options to disable fast-math optimizations:

• The precise qualifier can decorate any expression/variable to inhibit all fast-math operations
• The -Gis flag can be used to implicitly declare that all expressions are precise

Unfortunately, there is no "in between," meaning that you either have all of fast-math or none of it. In particular, being able to decouple the compiler setting for math associativity from math contraction optimizations would be a useful lever to pull in many shader programs.

Describe the solution you'd like
I would like the various fast-math optimizations to be enabled or disabled independently of each other.

Describe alternatives you've considered
Here, I describe the possibility of an intrinsic that suggests a contracted operation, even if precise or -Gis is requested.

Start by reviewing how the precise qualifier and -Gis flag currently disable fast-math optimizations, then read the linked hlsl-specs issue 883 for the related intrinsic proposal. Done should provide independent controls for associative expression rewriting, contraction, and other fast-math optimizations without requiring all or none of them.