make sure that only concat preallocates buffers

[ritchie46]

Which issue does this PR close?

Partial fix for #347. In #348 @jorgecarleitao pointed out that the memory savings don't work for the filter and the zip kernel.
This PR restores the implementation for those two kernels, and keeps the new preallocation for the concat kernel.

This is achieved by create a builder pattern for MutableArrayData. This way we don't break the API, and we may choose to preallocate buffers.

Could you take a look at this @jorgecarleitao ?

Are there any user-facing changes?

There is an builder pattern struct for MutableArrayData.

If there are any breaking changes to public APIs, please add the breaking change label.

No

[codecov-commenter]

Codecov Report

❌ Patch coverage is 59.21053% with 31 lines in your changes missing coverage. Please review.
✅ Project coverage is 82.58%. Comparing base (4ff2e56) to head (db0a5b4).

Files with missing lines	Patch %	Lines
arrow/src/array/transform/mod.rs	49.18%	31 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #382      +/-   ##
==========================================
- Coverage   82.61%   82.58%   -0.03%     
==========================================
  Files         162      162              
  Lines       44228    44329     +101     
==========================================
+ Hits        36538    36609      +71     
- Misses       7690     7720      +30     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jorgecarleitao]

Thanks @ritchie46 . My small concerns with this PR:

1. We are introducing yet another builder API for something relatively easy to accomplish with an extra method
2. preallocate_buffers assumes that "preallocate" means allocate the maximum possible capacity, which IMO is not really pre-allocating, is just being very conservative and allocating the maximum.
3. The API is rather limited to the concatenate case.

I left some ideas as a comment on the issue

[ritchie46]

@jorgecarleitao I implemented your proposal from #347. As we now need to define an enum in the with_capacities constructor, I want to be extra certain that I added all possible variants, as adding one later is backwards incompatable. Could you check if I missed something.

For now, only the (large)-utf8 ones uses this, but if the design is ok, I can add more in later PRs.

[alamb]

@jorgecarleitao what is the status of this one?

[jorgecarleitao]

Good to go: this is a solid improvement to me.

[alamb]

Included in #411 cherry pick