ARROW-11414: [Rust] Reduce copies in Schema::try_merge

[alamb]

I was looking at this code yesterday while using it in IOx -- https://github.com/influxdata/influxdb_iox/pull/703

Rationale:

Even though Schema::try_merge requires a slice of Schemas (not schema refs) ownership of its inputs, it copies all of its fields. This is inefficient ideal in the common case where most of the fields in the merged Schema will be the same

Changes:

This PR proposes to change the implementation so that try_merge takes something (like a Vec) that can iterate over the Schemas passed in and consume them, avoiding at least one copy per unique field. I intend no algorithmic changes, only performance improvement.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-11414

[alamb]

The change in this example shows a pretty good example of how I think the usability (as well as performance) of this API is improved by this PR

[jorgecarleitao]

LGTM. 👍

[alamb]

@houqp / @nevi-me -- any concerns about this PR?

(I think it was introduced here 494e7a9)

[houqp]

LGTM. In general, i think this is the pattern we would want to apply to the rest of the code base whenever applicable. i.e. if a function takes reference but does internal clones, then it's better to let caller manage the ownership instead.

[alamb]

i.e. if a function takes reference but does internal clones, then it's better to let caller manage the ownership instead.

Yes I like this approach a lot