sem: fix type inference for static parameters

[zerbina]

Summary

• fix compiler crash when passing empty aggregate values to static
  parameters
• fix static T parameter values not being of type T, when the
  argument is not directly of type T but requires a conversion
• fix converters not being considered for arguments to static
  parameters

Details

When matching an expression against a formal parameter that:

• is a static parameter
• contains a static type somewhere (e.g., int or static[float])

the expression was compile-time evaluated first, and on success,
assigned a tyStatic type. If the tyStatic argument type matched the
formal tyStatic type, the argument type was bound to the formal
tyStatic as-is.

In case an implicit conversion is necessary, the conversion was applied
only after the evaluated tyStatic was bound to the parameter type,
leading to:

• the actual value inlined at usages of the static parameter having
  the wrong type
• type inference for empty aggregate types not working

Solution

If the formal type is a static T, full argument matching (including
implicit conversion and converter handling) for between the argument
and T is performed first, and only in case of success is the
expression compile-time evaluated, and the resulting tyStatic type
bound to the parameter type.

Post-match fitting has to take place prior to const evaluation (in
order to correctly type empty aggregates), so a new procedure
tryEvalStaticArgument is introduced that handles the static
arguments.

The new static handling renders the macro/template static special-
casing obsolete; replacing the static argument with the evaluation
result is now done in evalTemplateArgs.

Typed AST

For calls to routines with static arguments, the typed argument
expression now stays in the AST as-is, instead of being replaced with
the evaluated value.

[zerbina]

The failure of tgeneric_various.nim is due to proc symbols passed to static[proc()] parameters now being turned into closures (as they should). The evaluated result is an nkClosure, however, which can currently not pass through sem again.

[zerbina]

For making static[proc()] parameters work again, I've added handling for nkClosure to semExpr -- nkClosure AST is now turned back into an nkHiddenStdConv. I think it's an okay solution for now; the alternatives would have been to either:

• implement full analysis for nkClosure
• OR turn closures without an environment into just an nkSym node in vmcompilerserdes

[saem]

I think the this solution, where semexprs handles nkClosure is better as it's less action at a distance (compared to vmserdes), and full nkClosure analysis would be built into sem, likely involving semexprs anyhow.

[saem]

/merge

[github-actions]

Merge requested by: @saem

Contents after the first section break of the PR description has been removed and preserved below:

---

Notes for Reviewers

• the typed argument expression staying in the typed AST makes sem: re-sem all call arguments in macro output  #1192 a bit easier