bug: -0.0 and 0.0 still compare unequal between columns and in array_has #9316

Description

#6236 (fixing #5868) makes filters treat -0.0 and 0.0 as the same value when one side is a literal. Comparisons between two float values, and array_has with a zero, still compare by encoding:

Steps to reproduce

import lance
import pyarrow as pa

data = pa.table({
    "id": [0, 1, 2],
    "a": [-0.0, 0.0, 1.0],
    "b": [0.0, -0.0, 1.0],
    "l": [[-0.0], [0.0], [1.0]],
})
ds = lance.write_dataset(data, "zero.lance", mode="overwrite")

def ids(filter):
    return sorted(ds.to_table(columns=["id"], filter=filter)["id"].to_pylist())

print("a = 0.0            ", ids("a = 0.0"))   # fixed by #6236
print("a = b              ", ids("a = b"))
print("a < b              ", ids("a < b"))
print("array_has(l, 0.0)  ", ids("array_has(l, 0.0)"))
print("array_has(l, -0.0) ", ids("array_has(l, -0.0)"))

On pylance 13.0.0b4:

a = 0.0             [0, 1]
a = b               [2]
a < b               [0]
array_has(l, 0.0)   [1]
array_has(l, -0.0)  [0]

Expected behavior

a = b               [0, 1, 2]
a < b               []
array_has(l, 0.0)   [0, 1]
array_has(l, -0.0)  [0, 1]

Lance version

13.0.0b4

Language binding

Python

Logs / traceback

• fix: make float filters treat -0.0 and 0.0 as the same value #6236's literal rewrite can't reach a column-to-column comparison. feat: Support IEEE 754 negative zero semantics apache/datafusion#22835 normalizes zeros in some DataFusion paths and IN LIST: treat signed zeros as equal apache/datafusion#25186 (open) does so for IN lists; the scan filter here still goes through Arrow's total-order kernels.
• array_has could be rewritten the same way as 0.0 IN (a, b) is in fix: make float filters treat -0.0 and 0.0 as the same value #6236: array_has(l, 0.0) → array_has(l, -0.0) OR array_has(l, 0.0).
• incorrect handling of -0.0 in comparison #5868 already points out that -1.0 * 0.0 produces -0.0 in ordinary analytics, which makes column-to-column cases like a * -1 = b realistic.