Defer dataclip search_vector indexing off the insert path

[stuartc]

Description

Same idea as #4818, but for dataclips instead of log_lines.

When a run finishes, the worker sends each step's output dataclip back to
Lightning to be saved. An AFTER INSERT trigger then builds the dataclip's
search vector with jsonb_to_tsvector. For big dataclips that's slow — slow
enough that under load the insert can sit on a connection past the ~70s timeout
and get rolled back. The dataclip never gets saved, and then everything after it
in the run (next step's input, log lines, final_dataclip_id) points at
something that doesn't exist, so the whole run is lost. That's one of the shapes
behind #4794.

So this takes the same approach we used for log lines: inserts now leave
search_vector NULL, and a background Oban worker
(Lightning.Invocation.DataclipSearchVectorWorker) backfills it out-of-band on
its own dataclip_search_indexing queue. There's a guarded
safe_jsonb_to_tsvector function and a partial index (WHERE search_vector IS NULL) so finding pending rows stays cheap. Dataclip search is now
eventually-consistent — usually caught up within a minute.

Three migrations: add safe_jsonb_to_tsvector, add the partial index, then drop
the trigger. dataclips isn't partitioned so the index is a single
CONCURRENTLY build (no per-partition dance like the log_lines one needed).

Caveat

This is one thing about this approach worth mentioning:

In the situation of an idle instace, with a single dataclip being inserted.
Indexing is poll-driven by the 1-minute cron - a lone insert can't enqueue the worker itself.
So with zero load, a new row waits for the next cron tick: up to ~60s, ~30s on average.
This is a floor set by the cron cadence, not by load.

Closes #4800

Validation steps

1. Run the migrations (mix ecto.migrate). Confirm the dataclips
set_search_vector trigger is gone, safe_jsonb_to_tsvector exists, and
   dataclips_pending_search_idx is VALID.
2. Insert a dataclip — its search_vector is NULL and it isn't matched by
   search yet.
3. Run Lightning.Invocation.DataclipSearchVectorWorker — the row's
   search_vector is populated and matches to_tsquery('english_nostop', …).
4. Insert one with an oversized body (>1MB tsvector) — it drains to an empty
   vector rather than erroring, and doesn't get stuck retrying.
5. mix test test/lightning/invocation/dataclip_search_vector_worker_test.exs test/lightning/runs_test.exs

Additional notes for the reviewer

1. The worker is basically the log_lines one — drains newest-first, snowballs an
   immediate follow-up while there's backlog, falls back to a 1-minute cron
   heartbeat. Concurrency 1 on its own queue. The snowball's unique states are
   restricted to [:available, :scheduled] on purpose (the default includes
   :executing/:completed, which makes a running job dedup its own successor
   and kills the chain).
2. Separate queue from log_lines' search_indexing on purpose, so a log-line
   backlog can't starve dataclip indexing (or vice versa).

AI Usage

• I have used Claude Code
• I have used another model
• I have not used AI

Pre-submission checklist

• I have performed an AI review of my code
• I have implemented and tested all related authorization policies.
• I have updated the changelog.
• I have ticked a box in "AI usage" in this PR

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.3%. Comparing base (9b000a0) to head (1aa5272).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@          Coverage Diff          @@
##            main   #4821   +/-   ##
=====================================
  Coverage   90.3%   90.3%           
=====================================
  Files        443     444    +1     
  Lines      22562   22577   +15     
=====================================
+ Hits       20379   20397   +18     
+ Misses      2183    2180    -3     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[github-actions]

Security Review ✅

• S0 (project scoping): N/A — the only data write is DataclipSearchVectorWorker updating dataclips.search_vector (lib/lightning/invocation/dataclip_search_vector_worker.ex:48-55), a system-level background index maintenance task whose SQL is intentionally cross-project; project scoping happens at the read side (Invocation.search_workorders).
• S1 (authorization): N/A — no new controllers, LiveView events, or web-layer entrypoints; the diff only adds Lightning.Config callbacks and threads them into an existing Oban worker.
• S2 (audit trail): N/A — no writes to configuration resources (workflows, credentials, project settings, etc.); the worker only backfills a derived search_vector tsvector column.