Quaternion: Return early from slerpFlat if t is 0 or 1

[jure]

Description

In Quaternion.slerp we return early if t is 0 or 1, returning qa if t 0 and qb if t 1. We don't do the same in slerpFlat and it impacts performance in certain cases.

1.000.000 slerpFlats without early return in msec (using the quats in in Quaternion.tests.js):

t	time in msec
0	56.661
1	60.9848
0.5	50.9151
0.25	47.1928
0.75	47.2224
0.5	36.2757
0.5	39.1083

1.000.000 slerpFlats with early return:

t	time in msec
0	12.8091
1	14.754
0.5	52.2216
0.25	48.1502
0.75	47.7769
0.5	34.5647
0.5	37.7248

There is some natural variation, but for t 0 and t 1 the early return variant is on average 4-5x faster.

In a real world example where slerpFlat is used to interpolate between player rotations in a multiplayer scenario, I'm definitely seeing an improvement.

Happy to code golf it a bit more, but I'm looking for feedback on whether the change is appropriate first.

[jure]

Anything else you'd like me to do here? Wondering so I can put a mental ✔️ here and let the winds of PR time take it away. It's not likely this will go out of sync with dev since it's a pretty stable piece of the codebase, so no hurry as far as I'm concerned.

[mrdoob]

Thanks!