fix(fast-path): recurse into child fields when computing next_fid

[FANNG1]

Fixes #41

Problem

FastPathFragmentWriter computed the field ID for new columns as:

next_fid = max(f.id() for f in self.lance_ds.lance_schema.fields()) + 1

lance_schema.fields() returns only top-level fields. A struct column with M children occupies M+1 consecutive Lance field IDs (1 for the parent + M for the children). The top-level-only scan misses the child IDs, so next_fid can collide with an existing child.

Example for a table with id (fid=0) and a struct meta (fid=1) with children meta.a (fid=2) and meta.b (fid=3):

Scan	max fid	next_fid	Effect
Top-level only	1	2	New file's "fields": [2] → Lance maps to meta.a
Recursive (fix)	3	4	Correct field ID for the new column

With the wrong next_fid, the committed fragment metadata maps the new file to the wrong schema field. Lance finds no file for the new column's actual field ID and returns null for every row.

The same problem occurs with any nested type: fixed_size_list, blob-v2, maps, etc.

Fix

Replace the flat max() with a recursive _max_field_id() that walks all descendants via f.children():

def _max_field_id(fields) -> int:
    best = -1
    for f in fields:
        best = max(best, f.id())
        children = f.children() if callable(getattr(f, "children", None)) else []
        if children:
            best = max(best, _max_field_id(children))
    return best

next_fid = _max_field_id(self.lance_ds.lance_schema.fields()) + 1

Test

TestRegressions::test_next_fid_skips_nested_child_field_ids — creates a table with a struct column (occupying 3 field IDs), merges a new scalar column via fast path, and asserts the column reads back as non-null with correct values.