FTS seems to always consider the first row even if it should be masked by prefilter

[westonpace]

Simple reproduction (courtesy of lancedb/lancedb#1656)

import lance
import pyarrow as pa

data = pa.table({
    "text": ["Frodo was a puppy", "There were several kittens playing", "Frodo was a happy puppy", "Frodo was a very happy puppy"],
    "sentiment": ["neutral", "neutral", "positive", "positive"]
})
ds = lance.write_dataset(data, "/tmp/test.lance", mode="overwrite")
ds.create_scalar_index("text", "INVERTED")
ds.create_scalar_index("sentiment", "BITMAP")

results = ds.to_table(full_text_query="puppy", filter="sentiment='positive'", prefilter=True, with_row_id=True)
print(results)
assert results.num_rows == 2

I suspect that the wand / posting iterator logic is doing something like (apologies in advance for my poor understanding of the wand search :) )...

candidate = iterator.current()
while not iterator.exhuasted():
  if candidate.matches_fts():
    iterator.advance_until_greater_than(candidate.score)

And the mask is only applied in iterator.next and so that first call to iterator.current() is always returning the first result, whether it matches the mask or not.