stats: prevent NaNs in histograms

[yuzefovich]

When running the reproduction from #141448

SET sql_safe_updates = false;

CREATE TABLE seed AS SELECT g AS _float8,
                            g * '1 day'::INTERVAL,
                            g % 2 = 1 AS _bool,
                            g::DECIMAL AS _decimal
                       FROM generate_series(1, 1) AS g;

INSERT INTO seed (_float8, _bool, _decimal) VALUES (0.9, true, 0);

UPDATE seed AS t SET _bool = t._bool AND false;

INSERT INTO seed (_bool, _decimal) VALUES (false, 1.0);

ANALYZE seed;

SHOW STATISTICS FOR TABLE seed;

SHOW HISTOGRAM <id>;

on M1 I'm getting

  upper_bound | range_rows | distinct_range_rows | equal_rows
--------------+------------+---------------------+-------------
  false       |          0 |                   0 |          0
  true        |          0 |                 NaN |          0

NaN is problematic since it then will prohibit the stats inclusion into the stmt bundle due to

stmtEnvCollector: json: unsupported value: NaN

I observed this behavior on two tables when looking into a stmt bundle issue on the customer cluster.

The root cause of the reproduction is being addressed elsewhere, but clearly we have some ways for NaNs to sneak into the histograms, so we should improve things.

Jira issue: CRDB-48121