Do not recost and rethread trees inside `optRemoveRangeCheck` (#69895)
* Do not set order in "optRemoveRangeCheck"
All but one caller of the method (RangeCheck) already do so in
their "for (GenTree* node : stmt->TreeList())" loops, so it is
not necessary.
Additionally, re-threading the statement, when combined with
"gtSetEvalOrder", can have the consequence of redirecting said
loops, possibly causing them to miss some trees, which was
observed in early propagation when working on removing "GT_INDEX".
A few small diffs because we no longer recost when removing
range checks in loop cloning, which is generally not necessary
because cloning runs before the "global" costing is performed,
except there is one quirk in "gtSetEvalOrder" which was looking
as "compCurStmt", and that is not set during "fgSetBlockOrder".
* Work around a TP regression
Gets us back 0.13% (!!!) instructions according to PIN.