output: optimize branch condition in metadata generation

[Watson1978]

Which issue(s) this PR fixes:
Fixes #

What this PR does / why we need it:
This PR optimizes the branch conditions inside the metadata generation in Fluent::Plugin::Output.

In most production use cases, either time or tag (or both) are used as chunk keys.

When running rake benchmark:run:in_tail with a 10 GB file:

Benchmark	before (sec)	after (sec)
rake benchmark:run:in_tail	41.220453	40.671516

Docs Changes:
N/A

Release Note:

• output: optimize branch condition in metadata generation

[kenhys]

How does it affect to <buffer []> case?

Before: [] => time,tag => time => tag
After: time,tag => time => tag => []

Because priority was changed in this case.

[Watson1978]

Yes, as you pointed out, the evaluation for the <buffer []> case is deferred to the end, so it is expected to be slightly slower than before.

Since the main goal of this PR is to optimize the most common cases (using time and/or tag), I believe this slight trade-off for the empty buffer case is acceptable.