Log Message:
imath: update to 3.1.2.

## Version 3.1.2 (July 31, 2021)

Patch release that fixes a Windows header issue.

* Improve handling of ``#include <*intrin.h>``

## Version 3.1.1 (July 20, 2021)

Patch release that fixes a build failure on ARM64 macOS

## Version 3.1.0 (July 13, 2021)

Minor release with new features:

* Optimized half-to-float and float-to-half conversion, using F16C SSE
  instruction set if available. Non-SSE conversion eliminates the
  float-to-half exponent lookup table, and half-to-float conversion
  provides a compile-time-optional bit shifting that is slower but
  eliminates the need for the lookup table, for applications where
  memory is limited.

  Half-to-float and float-to-half conversion is also available as
  C-language functions ``imath_half_to_float()`` and
  ``imath_float_to_half()``.

  All new conversions produced identical results, and new options are
  off by default to ensure backwards compatibility. See
  https://imath.readthedocs.io for more info.

* NOEXCEPT specifier can be eliminated at compile-time via the
  ``IMATH_USE_NOEXCEPT`` CMake option.

* Python bindings:
  * FixedArray objects support a "read only" state.
  * FixedArray objects support python buffer protocol.

* Optimized 4x4 matrix multiplication.