ESQL: replace Float/Double NaN with null

[costin]

Description

Floating number (Float and Double) provide a special, out of band value - NaN, Not-a-Number.

This occurs typically when dividing a number by zero; integer numbers don't have one and instead thrown an error.
To make things consistent, I propose we treat and convert NaN internally to null so it never reaches the user.
Several arguments to support this:

• JSON doesn't allow NaN (nor -Inf or +Inf for that matter)
• ES doesn't allow ingestion of non-finite double or float (regardless of whether JSON is used or not) - see NumberFieldMapper for floats and doubles.

[elasticsearchmachine]

Pinging @elastic/es-ql (Team:QL)

[elasticsearchmachine]

Pinging @elastic/elasticsearch-esql (:Query Languages/ES|QL)

[not-napoleon]

As I said at the meeting yesterday, I think it's less important which option we pick, so long as we pick one and do it consistently in all functions.

This has an advantage that it's easier to test - if nothing should return NaN, we can just add an assert to the unit tests that we never get back a NaN, rather than having to think about each case and if it should be NaN or null.

Would we also collapse the infinities into null?

[bpintea]

Just noting this for future lookups:

Would we also collapse the infinities into null?

We practically have:

elasticsearch/x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/expression/function/scalar/math/Pow.java

Lines 77 to 83 in 722853e

	private static double validateAsDouble(double base, double exponent) {
	double result = Math.pow(base, exponent);
	if (Double.isNaN(result) || Double.isInfinite(result)) {
	throw new ArithmeticException("invalid result when computing pow");
	}
	return result;
	}

[not-napoleon]

@bpintea Yes, that's correct. We decided that NaN, and both infinities should be rendered as null. I don't recall all the reasoning behind this, but one factor was that the infinite values aren't representable in JSON

[elasticsearchmachine]

Pinging @elastic/es-analytics-geo (Team:Analytics)

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)

[nik9000]

Save for a few bugs, I believe we've done this work.