Add volatile keyword to doubles involved in sorting operations for 32-bit GCC builds

[mckib2]

• resolves https://github.com/ERGO-Code/HiGHS/issues/769 and the associated SciPy issues

This is a well-known 32-bit GCC floating point precision "bug". This PR implements a minimal fix as presented in that thread (link to the specific comment).

__i386__ should only be defined by GCC for 32-bit systems as described here, so there should be no performance regressions or complications for 64-bit builds.