math_brute_force: Refactor input generation & reduce default test size

[rjodinchr]

This patch addresses execution time and input coverage in the
math_brute_force tests by centralizing input generation and scaling
down the default number of tested values.

Key changes:

• Abstracted input generation: Replaced duplicated array-filling logic
  across test files with shared fill*Input helper functions (for unary,
  binary, and ternary operations). This ensures that special edge-case
  values (NaNs, Infs, denormals) and random payloads are consistently
  injected across all tests.
• Reduced default test size: The default loop bounds have been reduced
  from an exhaustive 1ULL << 32 values to a randomized subset of
  1ULL << 27 inputs. This drastically speeds up standard test runs.
• Added exhaustive -a flag: Introduced a new -a command-line option
  that restores the exhaustive testing behavior, allowing validation
  against all 2^32 sequential values when needed.
• Improved wimpy mode scaling: Removed the scale and step skip logic,
  which could inadvertently skip combinations of special values. Instead,
  the patch now directly reduces the total input_count_power_of_two by
  using a De Bruijn sequence bit-hack to calculate the log2 of the
  wimpyReductionFactor.
• Consolidate the hardcoded arrays of edge-case test values from multiple
  individual test files into a single, shared location.

Ref #2669