ADBC driver for production in-process usage

[refset]

Opening this issue as a place to collect requirements for any possible future ADBC work.

Possible scenarios where ADBC could be useful:

1. Library similar to iceberg-core+iceberg-arrow, "can my data science team read the Arrow files in XT directly?" (secondarily worth being aware of this sort of extensibility: https://duckdb.org/2024/06/10/delta.html)
2. Maximum performance for running application code alongside an XT node (i.e. in-process without transport or serde overheads)

Reference implementation https://duckdb.org/docs/api/adbc

XTDB could be quite unique as transactional ADBC database that avoids roundtrip latencies with ~zero coordination/locking for long running reads.

[kpassapk]

A colleague asked this exact question yesterday: "Can I use the Arrow files directly? (From Spark)" Not a blocker for us, but ultimately useful for the data side of house to use XTDB more extensively.