Bignum: stop duplicating test function arguments for 32-bit vs 64-bit

[gilles-peskine-arm]

Some new bignum unit tests use functions that take arguments in pairs, with one argument used on 32-bit systems and one on 64-bit systems. This is useful because in some cases, the interesting inputs, or even some outputs such as carries, depend on the limb size. However, passing pairs of arguments in this way makes lists of test cases harder to read, complicates the test function code, and means that the same test case uses different data on different platforms.

The goal of this issue is to replace this idiom by one that does not involve passing different arguments depending on the limb size. Instead, if some test data is only valid for a given limb size, make the test case depend on MBEDTLS_HAVE_INTnn.

The test data generation framework (generate_bignum_tests.py) is now available to generate pairs of similar test cases in a convenient, easy-to-review way.

Definition of done: no more duplicated test arguments A4/A8, no more conditionals on defined(MBEDTLS_HAVE_INTnn) or sizeof(mbedtls_mpi_uint) in test_suite_mpi*.function.

Follow-up: #6371

[tom-cosgrove-arm]

Marking this as priority-medium as it should be tackled after the priority-high items. Note that already some of this is happening organically, so this will likely be a mop-up task anyway

[yanesca]

These same functions are not using the uniform test generation classes (which were added later). Doing the switch should include switching to the new python superclasses in test generation (OperationCommon and ModOperationCommon). Retaining identical test cases to the original is not a goal, should switch to the common inherited dataset.

Each function being reworked should be handled in a separate PR.

[yanesca]

Closing as more specific tasks have been raised to track this issue.