[XLA:GPU] Allow rewrites of the dot without non-contracting dims to m…

…ultiply+reduction if dot algorithm is F32_F32_F32

Such rewrite does not change the precision and makes this op 100x faster.

In the cl we:
1) convert the if block to the early exit version. (reduce indent)
2) relax the algorithm check by inverting the condition has_precision_config_algorithm ==> can_rewrite_dot_with_precision_config_algorithm (simplifies if-s)
3) add F32_F32_F32 algorithm to the can_rewrite_dot_with_precision_config_algorithm condition

PiperOrigin-RevId: 694491750