Simplify Zicsr AST

[nwf]

Split out from and sitting atop #617, the last commit here is the salient one. This transforms a bool * regidx under one ast data constructor into two ast data constructors, allowing us to stop punning between regidx and bits(5). This will be significant in light of #617 for supporting the E extension, where regidx will be (isomorphic to) bits(4); see #646.

[github-actions]

Test Results

396 tests  ±0   396 ✅ ±0   0s ⏱️ ±0s
  4 suites ±0     0 💤 ±0 
  1 files   ±0     0 ❌ ±0 

Results for commit 35eb511. ± Comparison against base commit 59c852c.

♻️ This comment has been updated with latest results.

[Timmmm]

Ah I understand the motivation now. In order to avoid losing the is_imm information could we use a union instead?

union RegOrImm = {
  Reg : regidx,
  Imm : bits(5),
}

[nwf]

I am... reasonably certain we haven't lost any information and that I haven't changed semantics. Rather than pass in a union and conditionally call X(), I just moved the discrimination into ast itself and had the appropriate arm unconditionally call X(). Can you say why you think a union is beneficial?

[arichardson]

I like the split into two separate AST clauses here but I think this changes the behaviour to no longer be correct.