4.x: RequestTracker exceptions can interrupt request lifecycle cleanup

[dkropachev]

Problem

Request-tracker callbacks run inline with request lifecycle transitions. A throwing tracker can prevent later cleanup from running, replace the original failure, or leave request state incomplete.

DefaultDriverContext returns a lone configured RequestTracker directly; only multiple trackers are wrapped by MultiplexingRequestTracker, which contains callback exceptions.

Examples include:

• CqlRequestHandler.handleRequestSetupFailure() calling onNodeError() before decrementing the active execution count;
• CqlRequestHandler.setFinalError() calling onError() before returning the throttler permit;
• equivalent unguarded tracker notifications in other success and failure paths.

Expected behavior

A RequestTracker exception must be logged and isolated. It must not:

• replace the request result;
• interrupt active-execution bookkeeping;
• skip timeout, stream, or throttler cleanup;
• affect another request.

Relationship to #1007

#1007 contains synchronous request-setup failures and performs their normal lifecycle cleanup. It intentionally does not add systemic RequestTracker exception isolation; doing that correctly requires auditing all tracker notification sites rather than guarding only the new setup-failure path.