producer: Batch commit requests to service.

Currently, we flush the buffer on every returned chunk or send-patches
request, which is causing unnecessary overhead. This change introduces a
delay period, during which commit requests are accumulated. When the
delay period ends, the buffer is flushed. The next time when a chunk
is returned or a patch request is sent, we schedule another delayed
flush. If the buffer grows too large or an immediate flush is requested
(via SharedMemoryArbiter::FlushPendingCommitDataRequests), we don't
wait for the delay period to end and flush immediately. This does not cancel any
pending delayed flushes.

Because batching can result in accumulated data being dropped at the end
of a tracing session, it has been made opt-in. Producers that want to
enable it should make sure that they enable asynchronous stopping of
data sources and that they call an immediate flush while the async stop
lasts.

Batching can also aggravate data loss during a producer crash, even with
scraping enabled, because the effectiveness of scraping is reduced by
unsent batched patches. This will be addressed in https://android-review.googlesource.com/c/platform/external/perfetto/+/1309414

This change is related to
https://chromium-review.googlesource.com/c/chromium/src/+/2106176,
https://chromium-review.googlesource.com/c/chromium/src/+/2149648 and https://android-review.googlesource.com/c/platform/external/perfetto/+/1309414

Bug: https://bugs.chromium.org/p/chromium/issues/detail?id=1029298
Change-Id: I7f8097e5b583412facc6b6d15aadc037a4fd6842