spirv-opt: Adding nested reassociation rules

[alister-chowdhury]

ReassociateNestedGenericInt
Reassociate integer instructions where both operands share the same opcode and both source instructions contain a constant.
e.g:

  (a * C0) * (C1 * b) = (C0 * C1) * (a * b)
  (a ^ C0) ^ (b ^ C1) = (C0 ^ C1) ^ (a ^ b)
  (C0 | a) | (b | C1) = (C0 | C1) | (a | b)
  (a & C0) & (b & C1) = (C0 & C1) & (a & b)

ReassociateNestedMulDivFloat
Reassociate floating point mul/div instructions, which have mul/div inputs, both of which contain a constant.
e.g:

  (a * C0) / (C1 / b) =  (C0 / C1) * (a * b)
  (C0 / a) * (b / C1) =  (C0 / C1) * (b / a)
  (a / C0) / (b * C1) =  (1 / (C0 * C1)) * (a / b)

ReassociateNestedAddSub
Reassociate add/sub instructions, which have add/sub inputs, both of which contain a constant.
e.g:

  (a + C0) - (C1 - b) =  (C0 - C1) + (a + b)
  (C0 - a) + (b - C1) =  (C0 - C1) + (b - a)
  (a - C0) - (b + C1) = (-C0 - C1) + (a - b)

[alister-chowdhury]

@luciechoi I'm not totally sure if the CI fails are related to my changes (but will happily fix them if they are).

CI-shaderc-smoketest

error: Operand '10[%10]' cannot be a type
  %13 = OpConstantSizeOfEXT %int %10

error: In Vulkan, untyped pointers can only be used in an explicitly laid out storage class
  %_ptr_UniformConstant = OpTypeUntypedPointerKHR UniformConstant

error: TypeBufferEXT must appear in a block
  %36 = OpTypeBufferEXT StorageBuffer

CI-macos-clang-release-bazel

//:opt_fold_test                                                        TIMEOUT in 60.1s
  /private/var/tmp/_bazel_kbuilder/2111b75ca87ac686a1e657996f9442aa/execroot/_main/bazel-out/darwin_x86_64-fastbuild/testlogs/opt_fold_test/test.log

[s-perron]

The failure in CI-macos-clang-release-bazel is a timeout in the folding tests.

Please resolve the conflicts, and we can look into the time out more. We might see if we can splitup the folding tests if it was not just a fluke.