fix(unittest): check not capturing operands

[zerbina]

Summary

Fix the value of some operand expressions to not, ==, etc. within
check calls not being captured, which would result in incorrect
check failure/success.

Details

Issue description

For displaying the value of the operands in case of a test failure,
the check macro "captures" the values of the operands prior to
evaluating the condition expression. This is implemented in
inspectArgs by assigning the operand expressions to temporaries and
then accessing the temporaries.

Type expressions cannot be assigned to run-time variables, so they were
guarded against, by testing for whether the nodes have ntyTypeDesc
as the type kind. However, since the AST passed to check is
untyped, this cannot work, but so far it did, due to an implementation
bug of typeKind, where the value returned for typeKind called on a
NimNode that has no type results in undefined behaviour.

In this specific case, the bug resulted in the notin {ntyTypeDesc}
expressions always to always evaluate to false, which in turn
disabled capturing for some expressions.

The solution

The decision of whether to assign the expression can only happen when
the expressions are typed. A two-phase macro could be used, but that
would require larger changes to check, so a different solution is
chosen: instead of an assignment, inspectArgs emits a call to the new
materialize template, passing along a generated nksTemplate symbol.
To access the materialized (or not) value, inspectArgs emits calls
with the generated template symbol.

When the materialize calls are later expanded, a template with the
passed-long symbol is produced. Using two materialize overloads, one
for typedescs and for everything else, doesn't work at the moment: if
the value expression is erroneous, no materialize template is
invoked, which would leave the gensym'ed symbol in an improper state,
later causing the compiler to crash when it attempts to use it during
overload resolution.

In addition, for improved clarity, inspectArgs is changed to use a
case statement.

[saem]

Nice catch.

wrt longer term solution, I think stuff like this should be more generics + comptime (similar to how one might do it in Zig) than macro heavy.

[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 change to the generated VM bytecode from make the in magic work the same across backends #949 break the notin {ntyTypeDesc} check
• prepares for fixing the typeKind bug