perf: Optimize round scalar performance

[kumarUjjawal]

Which issue does this PR close?

• Part of [EPIC] Optimize performance for slow expressions datafusion-comet#2986

Rationale for this change

The round function currently converts scalar inputs to arrays before processing, even when both value and decimal_places are scalar values. This adds unnecessary overhead for constant folding scenarios like

What changes are included in this PR?

• Add scalar fast path in RoundFunc::invoke_with_args for Float64 and Float32 inputs
• Directly compute the result when both inputs are scalars, avoiding array conversion overhead
• Add benchmark

Are these changes tested?

Yes

Type	Before	After	Speedup
round_f64_scalar	570 ns	195 ns	2.9x
round_f32_scalar	564 ns	192 ns	2.9x

Are there any user-facing changes?

No