ES|QL: Handle omitted exception message in Math.exact* calls

[GalLalouche]

As a follow up to #134910, we need to be more robust in the way we handle possibly null exception messages.

JDK 25 added optimizations that might cause the error message to be removed in some Math.exact* calls. See https://bugs.openjdk.org/browse/JDK-8367990.

Example of relevant (generated) code:

https://github.com/elastic/elasticsearch/blob/main/x-pack/plugin/esql/src/main/generated/org/elasticsearch/xpack/esql/expression/predicate/operator/arithmetic/SubIntsEvaluator.java#L112

Simplest solution I can think of is to a add a new kind of annotation with an extra error message on null, so the generated code would recreate the exception on null (possibly only when actually creating the exception in https://github.com/elastic/elasticsearch/blob/main/x-pack/plugin/esql/compute/src/main/java/org/elasticsearch/compute/operator/Warnings.java#L116).

[elasticsearchmachine]

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

[quackaplop]

JDK 25 added optimizations that might cause the error message to be removed
Thanks for cutting this! Just to clarify - this has always been allowed (the OmitStackTraceInFastThrow VM option controls this, off by default), but JDK25 seems to have made this optimization occur more often.

in terms if where to fix this, I agree that registerException is likely the right place.

Note that this affects other types of exceptions - see here - so it is worth checking that we do not rely on stacks/messages from these as a matter of correctness.