fix: reduce slice allocations in route dispatch

[alespour]

This PR makes handlers slice reused in router.matchAndDispatch in order to reduce slice (re)allocations. I think it is safe to reuse the slice since it is only used when order = true. Therefore, also added explicit if order check for iterating over handlers, although it is safe to keep iterating over nil slice when order = false, of course.

It is only a subtle optimization, but appeared when I profiled high-load scenario.

Benchmark with a single route:

before:

go test -v -run Benchmark_MatchAndDispatch -bench=. -benchmem -benchtime=10s

Benchmark_MatchAndDispatch-4   	13364338	       884.3 ns/op	     136 B/op	       3 allocs/op

after:

go test -v -run Benchmark_MatchAndDispatch -bench=. -benchmem -benchtime=10s

Benchmark_MatchAndDispatch-4   	14650513	       808.3 ns/op	     128 B/op	       2 allocs/op

I suppose with more routes, without reusing the slice, append would cause more re-allocations.

[MattBrittan]

Sorry for the long delay in responding to this, as you say it's a very subtile optimization but, I guess, the router forms part of the hot path so it's worth the slight decrease in readability. Thanks for the PR.