simplify(sessions): delete the test-only transcript upsert port

[ScriptedAlchemy]

What

Production transcript ingest only advances cursors through TranscriptWriteBatch::advance_offset; session and message rows arrive through observation admission and projection (since #2748/#2730). The whole Upsert write side was reachable only from tests, i.e. a test-only production port. This deletes it and migrates every test caller.

Deleted (no non-test callers, confirmed with rg plus an all-targets clippy build):

• TranscriptWriteBatch::upsert / upsert_with_cursor, TranscriptWriteKind (the batch now carries only the offset pair), and their unit tests
• the Upsert arm of GlobalDbTranscriptStore::persist_batch
• RegisteredGlobalDb::{upsert_transcript_batch, persist_transcript_batch_result, persist_transcript_batch_with_git_evidence_result} and the SessionStoreAccess chain behind them (upsert_transcript_batches_inner, batch message staging, write-time timestamp normalization, in-batch Codex goal reconciliation), TranscriptGitEvidence, TranscriptBatch
• TranscriptStoreError::{InvalidTranscriptPath, MissingTranscriptPath, MessageIdentityMismatch} and their ingest failure codes (only Upsert could produce them)
• test helpers upsert_transcript_batch_for_test (project test_support, dashboard test runtime), the unused global-db harness upsert_session_message_for_test, and set_parse_offset_insert_failure_for_test
• the stale merge_session_metadata comment

Migration

Fixture seeding goes through the existing primitives upsert_session + lcm_ingest_raw_message (no new port):

• seed_session_messages_for_test(scope, session, messages) -> Result<Vec<i64>> replaces upsert_transcript_batch_for_test (the dropped cursor key/offset args were never read back by any caller)
• upsert_session_message_for_test now returns Result<()> instead of Ok(false) on a swallowed write failure
• direct RegisteredGlobalDb callers (session-runtime lcm_effects, retained profile, temporal worker, daemon retention, automation fixtures) ingest raw messages directly

Tests deleted because they only protected the Upsert write: six session_suite::transcript_store full-batch tests (restart/replay, rollback, invalid batch, stale CAS, concurrent full batches, summary metadata survive-persist), the store-crate Upsert validation tests, the staging-attribution unit test, and the lcm_dag re-seed that asserted write-time microsecond-to-second normalization. The three advance_offset tests stay. The profile-scope Git-evidence refusal test now exercises the production span writer (record_span_observation) instead of the deleted transcript Git-evidence path.

Proof

This is a behavior-preserving deletion (simplify). It adds no behavior, so there is no new test that fails on master; the evidence is that every suite covering the remaining path passes.

Tests (non-zero counts):

• libs: tracedecay-store 72, tracedecay-sessions 522, tracedecay-session-memory 303, tracedecay-session-runtime 129, tracedecay-global-db 389 (+1 pre-existing failure, see below)
• tracedecay: session_suite 435, dashboard_api_test 78, automation_runner_test 171, transcript_ingest_suite 180, mcp_suite (git_correlation_test, mcp_handler_test::lcm_test) 30, hooks_lsp_suite hint_settlement 3, lib (daemon::scheduler, mcp::server::host_admission_tests) 43
• tracedecay-cli core_cli_suite (two migrated session tests) 2
• after rebase on fix(daemon): close stores in one teardown, drop capacity retry #2751: tracedecay lib 51, session_suite (lcm_/transcript_store) 186
• cargo clippy -p tracedecay-store -p tracedecay-sessions -p tracedecay-global-db -p tracedecay-session-memory -p tracedecay-project -p tracedecay-session-runtime -p tracedecay -p tracedecay-cli --all-targets --features ... -- -D warnings: clean
• cargo fmt --all -- --check: clean
• cargo check --workspace --all-targets --target x86_64-pc-windows-gnu --features tracedecay/test-transport,tracedecay/test-helpers,tracedecay-cli/test-transport: 0 errors (warnings are pre-existing cfg(windows) ones in untouched crates)

Pre-existing, not caused by this PR: tracedecay-global-db observation_collision_tests::vibe_jsonl_eof_refusal_survives_retention_generation_and_restart_without_rework fails identically on origin/master 411312b (same panic at observation_collision_tests.rs:2730), tracked in #2740.

Runtime journey on the built debug CLI (isolated HOME/TRACEDECAY_DATA_DIR, one daemon under systemd-run --user --scope -p MemoryMax=6G -p MemorySwapMax=1G, a Claude transcript bound to a scratch git project):

$ tracedecay init
initialized .../journey/project; daemon code-index reconciliation requested
$ tracedecay sessions import --project-path .
session import scheduled (session-sync.f82c16d6...); historical catch-up has remaining work 2
$ tracedecay sessions search "billing pipeline"
[claude] proj_1b1d853e1c2d8243 user: Investigate the billing pipeline regression
[claude] proj_1b1d853e1c2d8243 assistant: The billing pipeline regression is fixed.
# append a third message, re-import
$ tracedecay sessions search "invoice reconciliation"
[claude] proj_1b1d853e1c2d8243 user: Now audit the invoice reconciliation cursor
# re-import unchanged, then count all messages
$ tracedecay sessions search "journey OR billing OR invoice OR regression" | grep -c '^\[claude\]'
3

Ripwire: --quality-delta exit 0 (remaining rows are small constructor/wrapper shape clones and churn, no gating finding caused by this change); --edit-check on TranscriptWriteBatch, TranscriptWriteBatch::into_parts, TranscriptStoreError, persist_batch, upsert_session_message_for_test, seed_session_messages_for_test: incompatible=0.

[changeset-bot]

⚠️ No Changeset found

Latest commit: 75ffb4e

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR