sql/stats: add simple linear regression over float64s and quantiles

Statistics forecasting will initially use simple linear regression over
time to predict all table statistics (only keeping predictions that fit
the linear model well). Most table statistics are scalar float64s, for
which we can use a textbook ordinary least squares (OLS) method with x
as time and y as the table statistic.

For histograms, we use a variant of OLS based on the 2010 paper
"Ordinary Least Squares for Histogram Data Based on Wasserstein
Distance" by Verde and Irpino. The paper outlines an OLS method when
both x and y are histograms. In our case, x is a scalar, so we adjust
the math slightly.

Assists: #79872

Release note: None

pkg/sql/stats/BUILD.bazel

@@ -13,6 +13,7 @@ go_library(
         "new_stat.go",
         "quantile.go",
         "row_sampling.go",
+        "simple_linear_regression.go",
         "stats_cache.go",
     ],
     embed = [":stats_go_proto"],
@@ -71,6 +72,7 @@ go_test(
         "main_test.go",
         "quantile_test.go",
         "row_sampling_test.go",
+        "simple_linear_regression_test.go",
         "stats_cache_test.go",
     ],
     embed = [":stats"],