Make stack dumping errors display better.

Currently, the code first checks if a process is cached
and then checks if the process should be backtraced. For
example, the code should not backtrace the zygote process.
This CL moves the cached process check until after the
check to see if the process should be backtraced. Before
this change, the zygote was always in the list of processes,
and is always cached so it shows up as this anonymous
process that can't be backtraced.  After this change, since
the zygote is not supposed to be backtraced, it is not present.

For any pids that would have been dumped but are bypassed or have
an error, make the header include the timestamp and the command line.

Improve the message about why a cached process is skipped.

For an unwind that fails:
- If it's a java process, try a native unwind.
- Display the time it took for the failed unwind.
- Check if the process is now cached and display that error message.

Bug: 339473422

Test: Trigger bugreport and verify everything continues to work.
Test: Fake an unwind failure and verify the output.
Test: Fake an unwind failure and fake a cached process and verify the output.
Test: Fake an unwind failure for a java thread, and do a native unwind.
Test: Ran the bugreports from above through the bugreport parser.
Change-Id: I3082f2ac619e179621ad3ff39f2e839d000b2576