Add guard against invalid end timestamps

[JonasKunz]

What does this PR do?

Recently our windows tests have been failing:

2023-10-17T06:17:03.0410452Z 2023-10-17 06:16:57,817 [main] ERROR co.elastic.apm.agent.bci.IndyBootstrap - Advice threw an exception, this should never happen!
2023-10-17T06:17:03.0410818Z java.lang.AssertionError: [$.duration: must have a minimum value of 0]
2023-10-17T06:17:03.0411159Z JSON schema path = /apm-server-schema/current/transaction.json
2023-10-17T06:17:03.0411173Z 
2023-10-17T06:17:03.0411256Z {
2023-10-17T06:17:03.0411382Z   "timestamp" : 1697523417814567,
2023-10-17T06:17:03.0411487Z   "name" : "unnamed",
2023-10-17T06:17:03.0411602Z   "id" : "ddba5acdad68e7ab",
2023-10-17T06:17:03.0411791Z   "trace_id" : "054914ce0a91fb98a7e41a5ee516a021",
2023-10-17T06:17:03.0411943Z   "type" : "custom",
2023-10-17T06:17:03.0412102Z   "duration" : -5.49,
2023-10-17T06:17:03.0412207Z   "outcome" : "success",
2023-10-17T06:17:03.0412302Z   "context" : {
2023-10-17T06:17:03.0412465Z     "service" : {
2023-10-17T06:17:03.0412568Z       "framework" : {
2023-10-17T06:17:03.0412675Z         "name" : "API"
2023-10-17T06:17:03.0412761Z       },
2023-10-17T06:17:03.0412867Z       "version" : null
2023-10-17T06:17:03.0412952Z     },
2023-10-17T06:17:03.0413046Z     "tags" : { }
2023-10-17T06:17:03.0413134Z   },
2023-10-17T06:17:03.0413231Z   "span_count" : {
2023-10-17T06:17:03.0413337Z     "dropped" : 0,
2023-10-17T06:17:03.0413429Z     "started" : 0
2023-10-17T06:17:03.0413508Z   },
2023-10-17T06:17:03.0413637Z   "dropped_spans_stats" : [ ],
2023-10-17T06:17:03.0413738Z   "sample_rate" : 1.0,
2023-10-17T06:17:03.0413843Z   "sampled" : true
2023-10-17T06:17:03.0413923Z }
2023-10-17T06:17:03.0414510Z 	at co.elastic.apm.agent.MockReporter.verifyJsonSchema(MockReporter.java:406) ~[test-classes/:?]
2023-10-17T06:17:03.0415099Z 	at co.elastic.apm.agent.MockReporter.verifyJsonSchemas(MockReporter.java:398) ~[test-classes/:?]
2023-10-17T06:17:03.0415732Z 	at co.elastic.apm.agent.MockReporter.verifyTransactionSchema(MockReporter.java:338) ~[test-classes/:?]
2023-10-17T06:17:03.0416228Z 	at co.elastic.apm.agent.MockReporter.report(MockReporter.java:230) ~[test-classes/:?]
2023-10-17T06:17:03.0416769Z 	at co.elastic.apm.agent.impl.ElasticApmTracer.endTransaction(ElasticApmTracer.java:514) ~[classes/:?]
2023-10-17T06:17:03.0417282Z 	at co.elastic.apm.agent.impl.transaction.Transaction.afterEnd(Transaction.java:285) ~[classes/:?]
2023-10-17T06:17:03.0417763Z 	at co.elastic.apm.agent.impl.transaction.AbstractSpan.end(AbstractSpan.java:580) ~[classes/:?]
2023-10-17T06:17:03.0418251Z 	at co.elastic.apm.agent.impl.transaction.AbstractSpan.end(AbstractSpan.java:557) ~[classes/:?]
2023-10-17T06:17:03.0419074Z 	at co.elastic.apm.agent.pluginapi.AbstractSpanInstrumentation$EndInstrumentation$AdviceClass.end(AbstractSpanInstrumentation.java:223) ~[classes/:?]
2023-10-17T06:17:03.0419479Z 	at co.elastic.apm.api.AbstractSpanImpl.end(AbstractSpanImpl.java:126) ~[classes/:?]
2023-10-17T06:17:03.0420454Z 	at co.elastic.apm.api.BlockingQueueContextPropagationTest.testAsyncSpanDelegation(BlockingQueueContextPropagationTest.java:118) ~[test-classes/:?]

The problematic test manually sets the start-timestamp, but uses the agent built-in clock via end() for the end timestamp. Because those two methods could use two different "bases" for converting System.nanoTime() to micros, this potentially causes an end-timestamp which is lower than the start timestamp.

I assume we only observed this issue on windows because the precision of *nix System.currentTimeMillis() is higher and therefore makes this issue much less likely.

This PR adds a sanitization to the end-timestamp, as it is possible via manual instrumentation to give invalid timestamps.

Checklist

• This is an enhancement of existing features, or a new feature in existing plugins
  • I have updated CHANGELOG.asciidoc
  • I have added tests that prove my fix is effective or that my feature works
  • Added an API method or config option? Document in which version this will be introduced
  • I have made corresponding changes to the documentation

[github-actions]

/test

[SylvainJuge]

good catch !

[JonasKunz]

@SylvainJuge I've removed the test-on-windows label, because there are no other, new issues which are unrelated to this PR:
We are seeing several failures & flakes due to timeouts. I assume something must have changed with the underlying test infrastructure. I don't want this to block this PR, therefore I removed the label.

I'll come back to the windows tests next week and see if they are still having problems.