xorq stats: a stat taking min/max as float fails on int columns since D8, with no DAG warning #1051

Problem

Since dac617b (ADR-001 D8, in #1040), xorq min and max return ColumnValue and keep the column's type (xorq_stats_v2.py:164, :169). Before, they were cast to float64. A stat that declares them as float now gets an int on int columns, and the input check at stat_pipeline.py:121 rejects it because isinstance(1, float) is False. Float columns still work.

The DAG normally warns when a float consumer reads a ColumnValue provider. Here it doesn't: min and max are in XorqStatPipeline.EXTERNAL_KEYS (xorq_stat_pipeline.py:178), and build_typed_dag skips the type check for external keys (typed_dag.py:58). The first sign is a runtime error on each int column.

@stat()
def spread(min: float, max: float) -> float:
    return max - min

table = xo.memtable(pa.table({"i": [1, 5, 9], "f": [1.0, 5.0, 9.0]}), name="t")
sd, errs = XorqStatPipeline(XORQ_STATS_V2 + [spread], unit_test=False).process_table(table)
sd["f"]["spread"]  # 8.0
errs               # 'spread' expects 'min' as float, but got int: 1  (column i)

No stat in buckaroo, tallyman or pydata-app declares min/max as float. Tallyman project stats can't take them at all, since the loader builds a one-parameter compute(col). Only a stat written against the Python API is affected.

Suggested fix

Any of:

• A CHANGELOG note that min/max keep the column's type, so a stat reading them should declare ColumnValue or Any.
• Let the input check accept an int (not a bool) where float is declared, as type checkers do under PEP 484's numeric tower.
• Type-check an external key in build_typed_dag when a stat also provides it, so the mismatch warns at build time.

Context

Found during review of #1040. Line numbers are from the PR branch at a8df9b9.