sessions: retire test-only try_ingest_source driver and its parse_new replacement namespacing

[ScriptedAlchemy]

try_ingest_source (crates/tracedecay-sessions/src/runtime/source.rs) and the TranscriptSource::parse_new implementations it drives have no production caller: production catch-up is observation/port driven. Only test support (session_test_support.rs, transcript_ingest_suite/restart_atomicity.rs) reaches them. The Cursor and Codex parse_new paths still carry namespace_replacement_message_ids, a second message-id authority that production never uses.

After #2737 the scanner reports a replacement only for a genuine prefix rewrite, so this is not a live bug, but it is a parallel ingest authority kept alive by tests. Migrate those tests to the observation path, then delete the driver, the parse_new legacy paths and namespace_replacement_message_ids.

[ScriptedAlchemy]

Not closed by #2767. That PR fixes proportional JSONL resume after an earlier-byte edit, now tracked as #2762. It was requested under this number by mistake. This issue's acceptance (retire the test-only try_ingest_source driver and its parse_new replacement namespacing) is not done yet, so it stays open.

[ScriptedAlchemy]

claimed by fleet slug close-all-issues/verify

[ScriptedAlchemy]

Already done on origin/master. #2748 (2764eaa, 'delete the transcript parse/persist pipeline') deleted try_ingest_source, every TranscriptSource::parse_new implementation and namespace_replacement_message_ids; #2752 (e46da9e) deleted the remaining test-only transcript upsert port. Verified on origin/master fec824f: git grep -n 'try_ingest_source\|namespace_replacement_message_ids\|parse_new' over the whole tree returns no matches (the TranscriptSource trait itself remains, driven by observation admission). Tests that exercised host formats were migrated to the production admission path in #2748.