tp: fix flaky TempDir teardown crash in integration tests

[sashwinbalaji]

The perfetto_integrationtests binary was flakily aborting (SIGILL, exit
132) on sanitizer builds with:
temp_file.cc:154 PERFETTO_CHECK(Rmdir(path_)) (errno: 39, Directory not empty)

TempDir's destructor CHECKs that its directory is empty, so any leftover
file there takes down the whole binary. Two independent leaks caused this:

1. server unix registered its ".pid" path with the SIGTERM
   handler only after creating the file and setting up the listen socket
   and idle reaper. A SIGTERM in that window unlinked the socket but left
   the pid file behind, and ServerUnixBindAndCleanup then aborted on the
   non-empty TempDir. The socket is still removed, so the test's
   WaitForFileState guard passes and nothing looks wrong until teardown.
   The window is tiny normally but widens under ASan/MSan. Register the
   pid path before creating the file, mirroring the socket path.

2. Several tests removed a file from a TempDir only at the end of the test
   body, which is skipped if an ASSERT above returns early. Move those to
   OnScopeExit so cleanup always runs (traced multi-producer-socket test
   and the trace_processor bundle tests).

https://github.com/google/perfetto/actions/runs/28811487303/job/85440268082?pr=6547

Bug: 530236105

[github-actions]

🎨 Perfetto UI Builds

• d050dbb (2026-07-06 17:48 UTC): https://storage.googleapis.com/perfetto-ci-artifacts/gh-28811487303-1-ui/ui/index.html
• 21d4a5a (2026-07-06 18:47 UTC): https://storage.googleapis.com/perfetto-ci-artifacts/gh-28815058111-1-ui/ui/index.html