Replace some Sprintfs with string concatenation

[dnr]

What was changed

Replace Sprintf with concatenation in a few places.

Why?

This was visible in a profile of some worker pods doing heavy replays so I thought I'd try to improve it.

I created a benchmark out of the schedules replay test:

before:
     272          21243887 ns/op         4882488 B/op      91354 allocs/op
after:
     278          20966780 ns/op         4862933 B/op      90165 allocs/op

(best of 10 runs for each, after had less variance also)

So a 1-2% improvement in cpu and a tiny bit less memory.

Checklist

2. How was this tested:
   existing unit tests, plus simple benchmark

[cretz]

Not sure this is worth it, but made a comment where maybe simple string concatenation can be good enough for you

[cretz]

Suggested change

	state.yield(printfString{fmt: "blocked on %s.Receive", arg1: c.name})
	state.yield("blocked on "+c.name+".Receive")

Can you microbenchmark this approach?

[dnr]

Ah, yeah, that provides a basically identical improvement, and is simpler. I'll switch to that. I might have hoped the compiler could optimize this but I guess Go doesn't go for stuff like that

[Quinn-With-Two-Ns]

Are you running with PGO during your benchmark?

[dnr]

No, I don't think I've ever built a server with PGO. That's an interesting angle to try. I couldn't find any indication that the compiler would partially-evaluate a fmt.Sprintf even with PGO though.

(In general we don't care much about cpu usage since the server is generally io-bound, but in this situation it was replaying a ton of schedule workflows and going over its cpu reservation, so I looked for some quick improvements there.)