semfold: fix range-check folding (#822)

## Summary

Fix literals resulting from folding range-check conversions having a
type that doesn't match the node kind. This, in turn, fixes a VM code
generator issue where the code generated for an access of the result of
an integer-literal-converted-to-float-range would cause the VM to crash.

## Details

The folding logic directly assigned the destination type to the literal,
which led to, for example, integer literals having a `tyFloat` type.
This breaks the assumption of only float-literal nodes having a float
type, and is what caused the `vmgen` bug (`vmgen` only looks at the node
kind).

Same as for normal conversion, range-check conversion are now also
folded via `foldConv` (but only in case the range check succeeds),
fixing the issue.

compiler/sem/semfold.nim

@@ -761,8 +761,8 @@ proc getConstExpr(m: PSym, n: PNode; idgen: IdGenerator; g: ModuleGraph): PNode
     let a = getConstExpr(m, n[0], idgen, g)
     if a == nil: return
     if leValueConv(n[1], a) and leValueConv(a, n[2]):
-      result = a              # a <= x and x <= b
-      result.typ = n.typ
+      # a <= x and x <= b
+      result = foldConv(n, a, idgen, g, check=false)
     else:
       result = g.config.newError(n,
                                  PAstDiag(kind: adSemInvalidRangeConversion))

tests/lang_types/range/trange.nim

@@ -1,4 +1,5 @@
 discard """
+  targets: "c js vm"
   output: '''
 TSubRange: 5 from 1 to 10
 #FF3722
@@ -154,3 +155,11 @@ block:
   c = d
   doAssert a == b
   doAssert c == d
+
+block int_literal_to_float_range:
+  type Range = range[0'f..1'f]
+
+  var val = 0'f
+  # use the value in a comparison in order to make sure that
+  # the conversion really worked
+  doAssert val == Range(0)