add exhaustive AArch64 decoder/encoder/macro consistency tests

[derekbruening]

For x86 we have exhaustive tests of every INSTR_CREATE and XINST_CREATE
macro with self-consistency tests that the decoder and encoder agree on
everything.

For ARM we are still missing this: #1686.

This issue covers adding such tests for AArch64.

Without these tests, it is fragile to add new XINST_CREATE macros: we have
no idea whether they work.

[derekbruening]

Xref #2440

[egrimley]

Do the tests of XINST_CREATE macros have to be architecture-specific? It might be better if they weren't. At the moment it is perhaps not always clear what kinds of operand are allowed (register or immediate), what range of immediate values, and so on.

Note that we already have "api.dis-a64" for testing the A64 encoder and decoder, without the macros.

[derekbruening]

Sure, some of the XINST could be shared.

The api.dis test is a decoding and disassembly test, comparing the DR decoder/disassembler to other disassemblers using a moderately large, relatively random binary input and comparing the output to what other tools agree on. At least that's what it is for both x86 and ARM.

The api.ir test that this issue covers is an internal self-consistency test that systematically goes through every INSTR_CREATE macro with every type variation and ensures it can encode and be decoded back and that all varieties of decoder agree (fast and slow). It says nothing about agreeing with a ground truth: only internal consistency, but it systematically covers the full IR, whereas the decode-this-binary api.dis test only covers what happens to be in the binary input.

[egrimley]

The test "api.dis-a64" has a big set of cases (dis-a64.txt) and checks encoder and decoder (dis-a64.c). It doesn't check different varieties of decoder because those aren't implemented yet (i#2374).