traced_probes: Refactor + Remove timeout from aflags invocation (#5514)

- Remove the timeout on flush: this is unnecessary and
  adds just another timing parameter that makes the code
  harder to reason about. If aflags takes too long, the
  existing code in both TracingServiceImpl (5s) and
  ProbesProducer (1s) will take care of it.
- Unify some of the code paths that handle error cases.
- Added Subprocess::Kill (without Wait). This is to avoid
  suspending traced_probes' main thread waiting for the
  termination of a process in case of timeouts: if aflags
  takes too long, the trace will be finalized, hence the
  AndroidAflagsDataSource will be destroyed. By default
  the Subprocess dtor does a KillAndWait (and cannot
  do otherwise to avoid zombie processes). So we try
  to reduce the chance of waiting by allowing up to
  10 seconds between the kill and the waitpid().

Bug: 502852101