explain_plan does not accurately represent projection of nested subfields

Using this script

import lancedb
import random
from pydantic import BaseModel
from lancedb.pydantic import LanceModel

class Document(BaseModel):
    content: str
    source: str

class NestedSchema(LanceModel):
    id: str
    doc: Document

# this works
db = lancedb.connect(
        "my-db",
)

data = [{
    "id": "foo",
    "doc": {
        "content": "foo",
        "source": "foobar",
    },
}]

key = random.randint(1, 1000000)
table = db.create_table(f"test{key}", schema=NestedSchema, data=data)
print(table.search().where("`doc`.`source` = 'foobar' and id <> ''").explain_plan())

I get this plan:

ProjectionExec: expr=[id@0 as id, doc@1 as doc]
  Take: columns="id, (doc), _rowid"
    CoalesceBatchesExec: target_batch_size=1024
      FilterExec: get_field(doc@1, source) = foobar AND id@0 != 
        LanceScan: uri=my-db/test919631.lance/data, projection=[id, doc], row_id=true, row_addr=false, ordered=true

From the plan, it's unclear why "Take" is required at all, because [id, doc] are projected from the LanceScan. The reason is that only the "source" subfield of doc is really getting projected. We should update the explain_plan representation to indicate [id, doc@1] instead of what it's currently doing.

Start at the explain_plan entry point and reproduce the nested-field query from the issue using the provided LanceDB and Pydantic script. Trace how the projection is represented in the plan, then verify that the output distinguishes the projected nested subfield from the full doc field, such as showing id and doc@1 rather than only id and doc.