Add linear regression aggregation functions (regr_slope, regr_intercept, regr_r2, regr_count)

Description

I'd like to propose adding the core SQL-standard linear regression aggregation functions to Polars:

• pl.regr_slope(y, x) — slope of the least-squares linear fit of y on x
• pl.regr_intercept(y, x) — intercept of the fit
• pl.regr_r2(y, x) — coefficient of determination
• pl.regr_count(y, x) — number of rows where both y and x are non-null

These are standard aggregates in PostgreSQL, Spark SQL, DuckDB and DataFusion, with SQL-standard argument order (dependent variable y first) and semantics: null pairs are excluded pairwise; zero variance in x yields null for slope/intercept/r2; zero variance in y alone yields r2 = 1.

df.group_by("g").agg(
    pl.regr_slope("y", "x"),
    pl.regr_intercept("y", "x"),
    pl.regr_r2("y", "x"),
)

# and via the SQL interface:
df.sql("SELECT g, REGR_SLOPE(y, x) FROM self GROUP BY g")

Why it fits well into Polars:

• Polars already has pl.corr / pl.cov, but nothing regression-shaped; today users have to compose cov(x, y) / var(x) manually (and the intercept/r2 variants are even more verbose), which is error-prone around null handling.
• The implementation can reuse the numerically stable PearsonState moment state that already exists in polars-compute (weight, mean_x, mean_y, dp_xx, dp_xy, dp_yy) — that state is exactly the sufficient statistic for all of these aggregates, so no new kernels are needed and the streaming engine gets full support (including streaming group-by) via a GroupedReduction analogous to the existing Pearson one.
• The SQL interface gets REGR_SLOPE / REGR_INTERCEPT / REGR_R2 / REGR_COUNT for free, improving PostgreSQL compatibility.

The full family in PostgreSQL also includes regr_avgx, regr_avgy, regr_sxx, regr_syy, regr_sxy; I'd suggest starting with the four above and following up with the rest if there is interest, since they share the same machinery.

I have a working implementation (both engines, SQL, tests validated against DuckDB) and would be happy to open a PR.