Optimize array_join by supporting PROVIDED_BLOCKBUILDER convention

[wenleix]

Please make sure your submission complies with our Development, Formatting, and Commit Message guidelines.

Fill in the release notes towards the bottom of the PR description.
See Release Notes Guidelines for details.

== RELEASE NOTES ==

General Changes
* Optimizer performance for array_join

[linux-foundation-easycla]

The committers are authorized under a signed CLA.

• ✅ Wenlei Xie (f6ebeda52f2bec4a7519a847f2eb081ad665c9d1)

[wenleix]

Benchmark shows over 10% improvements.

Before

Benchmark                     Mode  Cnt    Score   Error  Units
BenchmarkArrayJoin.benchmark  avgt   60  152.954 ± 1.246  ns/op

After

Benchmark                     Mode  Cnt    Score   Error  Units
BenchmarkArrayJoin.benchmark  avgt   60  134.558 ± 2.078  ns/op

[wenleix]

See #9638 and #12166 for context.

cc @oerling , we once talked about allowing scalar function to directly write to output buffer (to avoid copy data for struct types). The framework is implemented but not function is yet using it. @kaikalur recently also observed such inefficiency when optimizing user's query, so here is an example about how to use it :) .

ArrayJoin only gets moderate benefit as the function logic is also quite intensive. This type of optimization would have more improvements for functions with light computations :)

[wenleix]

I realized the benchmark is over ARRAY(BIGINT) thus casting from BIGINT to VARCHAR can take significant time. Benchmark over ARRAY(VARCHAR) would probably show more improvements :)

[highker]

lgtm