refactor: introduce SchemaAdapter to perform logical/physical transform

[Xuanwo]

Close #4595

This PR introduces a SchemaAdapter to handle logical and physical transformations.

I’m wondering if this is the right place, or if we need a proper logical type system. I recall @westonpace saying we should remove the logical type concept entirely. So what’s our plan for handling conversions like jsonb to json or large_string to binary?

---

This PR was primarily authored with Codex using GPT-5-Codex and then hand-reviewed by me. I AM responsible for every change made in this PR. I aimed to keep it aligned with our goals, though I may have missed minor issues. Please flag anything that feels off, I'll fix it quickly.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 85.39326% with 13 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.69%. Comparing base (e3cf56a) to head (88e4cc9).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
rust/lance/src/dataset/utils.rs	84.14%	10 Missing and 3 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5096      +/-   ##
==========================================
- Coverage   81.72%   81.69%   -0.04%     
==========================================
  Files         340      340              
  Lines      138875   138886      +11     
  Branches   138875   138886      +11     
==========================================
- Hits       113498   113461      -37     
- Misses      21636    21685      +49     
+ Partials     3741     3740       -1     

Flag	Coverage Δ
unittests	81.69% <85.39%> (-0.04%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.

[westonpace]

This is fine. In a perfect world I think maybe we would have something like...

fn to_physical_stream(&self, stream: SendableRecordBatchStream) -> PhysicalRecordBatchStream;

This way the compiler could make sure that we aren't accidentally using the logical stream in places we should be using the physical stream (or vice-versa).

However, we can think about this as a todo for now.