A few fixes when dumping long running traces

This CL has a few fixes:
1. When recording on stack methods at the start of the trace, the
   current buffer index wasn't set correctly. It was set to the buffer
   size which basically means there are no entries. Fixed by setting it
   to the correct index.
2. We have some placeholder entries which should be ignored when dumping
   the trace.
3. We should encode the eventy type as a first value. Based on the event
   type we read the next value only for method entry events.
4. Encode the number of bytes written for each block of events, so it is
   easy when parsing.
5. Don't remove the event type when recording the timestamp_and_action.
   We need to clear the lsb to get the timestamp but when writing the
   entry we encode both timestamp and the event type as a single value.

Bug: 352518093
Test: Manual testing

Change-Id: Ibe72ad82daa37eb2f4760decbb207663836796c7

0 files changed, 0 insertions, 0 deletions