Fix divWithIntCheck for floats larger than max int

[rotu]

The integer accuracy fix could previously return a wrong result when its operands were bigger than max int.

[fuzzie360]

Closing as superseded: the divWithIntCheck rewrite in 2.19.0 (for #742) removed the int() casts entirely — the divisibility check now rounds the quotient and verifies by exact float multiplication, so the out-of-range undefined behavior this PR targeted no longer exists. (As a side note, int(x) == x would not compile on GLSL ES 1.00 — no implicit int→float comparison — so this needed a different shape anyway.) Thanks for flagging the UB.