[r] Enhance support for (partial) table extraction

[eddelbuettel]

This PR adds support for access via the SOMAReader class to extract selected columns in a single pass. Query condition and range support will be added next.

The retrieved data structure is a from an Arrow-support package that works only on the lighterweight C interface to Arrow without linking. Once data has been accessed, we can easily transfer to different Arrow data structures:

> library(tiledbsoma)
> uri <- "test/soco/pbmc3k_processed/obs"  # local data set
> columns <- c("n_counts", "n_genes", "louvain")
> z <- export_recordbatch(uri, columns)
[2022-10-07 18:00:12.785] [tiledbsoma] [Process: 66049] [Thread: 66049] [info] Reading from test/soco/pbmc3k_processed/obs
[2022-10-07 18:00:12.795] [tiledbsoma] [Process: 66049] [Thread: 66049] [info] Read complete with 2638 obs and 3 cols
[2022-10-07 18:00:12.795] [tiledbsoma] [Process: 66049] [Thread: 66049] [info] Accessing n_counts at 0
[2022-10-07 18:00:12.795] [tiledbsoma] [Process: 66049] [Thread: 66049] [info] Accessing n_genes at 1
[2022-10-07 18:00:12.795] [tiledbsoma] [Process: 66049] [Thread: 66049] [info] Accessing louvain at 2
> 
> rb <- arch::from_arch_array(z, arrow::RecordBatch)   ## Arrow RecordBatch
> rb
RecordBatch
2638 rows x 3 columns
$ <float not null>
$ <int64 not null>
$ <large_binary not null>
> 
> tb <- arrow::as_arrow_table(arch::from_arch_array(z, arrow::RecordBatch))  ## Arrow Table
> tb
Table
2638 rows x 3 columns
$ <float not null>
$ <int64 not null>
$ <large_binary not null>
> 

[shortcut-integration]

This pull request has been linked to Shortcut Story #22234: Improved tiledb-soma R ops.

[eddelbuettel]

(rebased and force-pushed)

[johnkerl]

LGTM