fix: resolve type aliases when lowering 'is' to Lua (#1149)

[chiliec]

What

Closes #1149.

x is SomeType compiles to a runtime type check. When the tested type is a type alias, the code generator inspected the nominal type's own typename ("nominal") instead of the type it resolves to, so integer and nil aliases were lowered like a generic type(x) == "..." check.

local x: any = nil
local type A = integer
print(x is A)   -- was: type(x) == "number"   (wrong)
local type B = nil
print(x is B)   -- was: type(x) == "nil"      (wrong)

x is A should lower to math.type(x) == "integer" (a number may be a float, not an integer), and x is B to x == nil — exactly what x is integer / x is nil already produce.

Fix

In both is-lowering sites (teal/gen/lua_generator.tl and the 5.1/5.3 compat header selection in teal/gen/lua_compat.tl), resolve the cast type through its nominal chain before discriminating on integer / nil. Chained aliases (local type C = A) resolve too.

Tests

Added a util.gen regression test in spec/lang/operator/is_spec.lua asserting alias and direct forms lower identically.

Verified genuine RED→GREEN: reverting the source fix (keeping the test) makes it fail with type(x) == "number" vs the expected math.type(x) == "integer"; restoring it passes.

Validation

make selfbuild            # two-stage bootstrap diff clean
busted spec/lang          # 1733 successes / 0 failures
busted spec/api           # 96 successes / 0 failures
busted spec/cli           # 112 successes / 0 failures

Regenerated .lua artifacts via make selfbuild so the committed build matches the .tl sources. Happy to adjust.

[hishamhm]

Thank you!