further performance improvements

[anzdaddy]

• make ci for consistent builds in CI.
• Adaptively run 386 builds in docker on laptop and bare in CI.
• More flexible profiling in Makefile.
• Detect memory allocation regressions.
• Rework numerous functions to reduce copying.
• Remove several unused functions.
• Improve constant array lookups.
  • Use arr[i%len(arr)] to avoid bounds checks.
  • Pad to 2ⁿ to avoid weird i%len(arr) math.
• Shrink some types for better packing.
• Change flavor constants for more efficient bit calcs.
• Optimise 128-bit divide by 10¹⁵ and 10¹⁶.
• Optimise all uint128T operations to minimise copying.
• Simplify and optimiser decimal unpacking. Split flavor unpack into a separate operation for earlier short-circuiting.
• Remove original field from decParts.
• Add replayOnFail for easier test replay.
• Rename IO benchmarks for better grouping.
• Faster Add, Cmp and Mul.
• Much faster sqrt using table lookup.
  • Add squareroot.decTest suite.