feat(python): Support shortcut eval of common boolean filters in SQL interface "WHERE" clause

[alexander-beedie]

Closes #18373.

Update

• Shortcut-eval for common SQL "WHERE" clause filters such as:
  • WHERE TRUE
  • WHERE FALSE
  • WHERE 1 = 1
  • WHERE 1 != 0

This pattern is a bit odd-looking, but often arises from programmatic query builders that want to append to a WHERE clause without having to check if a new filter is the first (so they can always "AND" it onto the existing clause). As a result a query may always contain "WHERE 1=1" (or equivalent), with the expectation that it is optimised out.

When the SQL parser sees that a WHERE clause represents such a pattern it now skips frame filtering entirely, immediately returning the same frame (if TRUE) or an empty frame with an identical schema (if FALSE).

(Note: you usually see the "TRUE" version of this, but once in a blue moon you'll see something evaluating to "FALSE" as some query-building optimisers may generate this if they can evaluate the truth-value of a constraint clause before it hits the backend).

Also

Minor internal module cleanup/refactor:

• Datatype-specific code moved from sql_expr.rs to new types.rs.

Examples

import polars as pl

df = pl.DataFrame({
  "x": ["aa", "bb", "cc"],
  "y": [1, 2, 3],
})

df.sql("SELECT * FROM self WHERE TRUE")
# shape: (3, 2)
# ┌─────┬─────┐
# │ x   ┆ y   │
# │ --- ┆ --- │
# │ str ┆ i64 │
# ╞═════╪═════╡
# │ aa  ┆ 1   │
# │ bb  ┆ 2   │
# │ cc  ┆ 3   │
# └─────┴─────┘

df.sql("SELECT * FROM self WHERE 1 != 1")
# shape: (0, 2)
# ┌─────┬─────┐
# │ x   ┆ y   │
# │ --- ┆ --- │
# │ str ┆ i64 │
# ╞═════╪═════╡
# └─────┴─────┘