Reject oversized GROUPING masks during analysis #26135

Describe the bug

GROUPING can panic during analysis of GROUPING SETS with 65 distinct grouping columns, even when GROUPING has only one argument.

The analyzer constructs the mask using 1 << group_by_idx. A column at bit 64 causes a shift overflow in a checked build. The physical grouping-set representation already has a 64-bit capacity limit.

To reproduce

Generate the SQL below and run its output in a debug DataFusion CLI:

n = 65
cols = ", ".join(f"c{i} INTEGER" for i in range(n))
keys = ", ".join(f"c{i}" for i in range(n))
row = ", ".join(str(i + 1) for i in range(n))
print(f"CREATE TABLE wide_keys ({cols});")
print(f"INSERT INTO wide_keys VALUES ({row});")
print(f"SELECT GROUPING(c0), COUNT(*) FROM wide_keys "
      f"GROUP BY GROUPING SETS (({keys}), ());")

Expected behavior

Return NotImplemented before constructing an unrepresentable grouping-set mask. Ordinary GROUP BY with 65 columns should continue to allow GROUPING to return zero.

Additional context

Reproduced through core API regression tests on commit 1fa378d380fa7e494aae1085945f8b048e9376c3.