Desired-balance warn threshold logging should accumulate across restarts

[DaveCTurner]

Today we emit periodic INFO logs about an ongoing desired balance computation which has not converged after some amount of time or some number of iterations:

elasticsearch/server/src/main/java/org/elasticsearch/cluster/routing/allocation/allocator/DesiredBalanceComputer.java

Lines 321 to 329 in 18f960c

	logger.log(
	reportByIterationCount || reportByTime ? Level.INFO : i % 100 == 0 ? Level.DEBUG : Level.TRACE,
	() -> Strings.format(
	"Desired balance computation for [%d] is still not converged after [%s] and [%d] iterations",
	desiredBalanceInput.index(),
	TimeValue.timeValueMillis(currentTime - computationStartedTime).toString(),
	iterations
	)
	);

However these numbers reset if a new cluster state is received, so it's possible for a steady stream of cluster states to prevent the computation for converging without ever seeing any warnings. IMO we should let these numbers accumulate until the computation fully converges, and report the number of restarts since convergence in the log message too.

[elasticsearchmachine]

Pinging @elastic/es-distributed (Team:Distributed)

[idegtiarenko]

Related to #91386

[elasticsearchmachine]

Pinging @elastic/es-distributed-obsolete (Team:Distributed (Obsolete))

[elasticsearchmachine]

Pinging @elastic/es-distributed-coordination (Team:Distributed Coordination)