graph build crashes on its first statement: the module forwarders were deleted instead of fixed

[cdeust]

Symptom

On 3.1.0 the graph never builds. /api/graph/progress reports
phase: "starting", pct: 0.0 and never advances; Graph, Brain, Knowledge, Board and Wiki stay
empty. Trace is unaffected — it does not go through this builder. Nothing is lost or corrupted:
the build dies before reading anything.

[cortex] background build error: module 'cortex_viz.server.graph_event_stream' has no attribute 'reset'
  File "cortex_viz/server/graph_build_run.py", line 155, in run_build
    _events.reset()
AttributeError

Root cause — and what it is not

45d4a80 deleted the module-level emit/close/reset forwarders from graph_event_stream,
justified in the code as: they "had never had a caller in this repository's history".

That premise is false. They had four:

caller	call
graph_build_run.py:155	_events.reset()
graph_build_run.py:234	_events.emit(...)
graph_build_run.py finally	_ev.close()
graph_build_merge.py:136	events.emit(...) — handed the module, per its own docstring

The commit's other stated reason is the real defect: emit grew an event_meta parameter and the
forwarder silently dropped it. That is a bug in the forwarder, and the reason to fix it. It was
used instead as the reason to delete it — so the root cause (a forwarder that had drifted out of
sync with the method it forwards to) was never addressed, and the deletion took four working
callers down with it.

The module is also published API: dropping three public functions in a patch-level move breaks any
consumer outside this repository, not just these callers.

What the fix must do

Restore the three forwarders plugged and functional — not route the callers around them:

• emit(label, nodes, edges, *, chunk=1000, event_meta=None) — forwards every parameter,
  event_meta included. The drift that motivated the deletion is what gets fixed here.
• close() and reset() restored, delegating to the singleton.
• graph_build_run.py and graph_build_merge.py keep calling the module — unchanged.
• Remove the "there are deliberately no module-level forwarders" comment; it now documents a
  decision that was withdrawn.

Acceptance criteria

• A test asserts each forwarder delegates to the process singleton returned by get_stream().
• A test asserts emit's forwarder carries event_meta through to what a subscriber receives —
  the exact drift that started this. It must fail against the pre-45d4a80 forwarder.
• A test exercises the binding the production build actually uses, so a call site and the module
  cannot drift apart again. Every existing merge test injects a fake stream, which is why none of
  them caught this: the defect was in how the caller obtains the stream, not in what it does
  with it.
• End-to-end: the standalone server advances past starting and reaches layout bake, with zero
  background build error in the log.

Notes

Verified locally that restoring the call path makes the build run to completion (baseline_ready,
85 hooks, 49 discussion_agents, 36752 memories, then layout bake (DrL)). PR #133 took the wrong
route — it adapted the callers to the deletion, which dresses a removal up as a solution — and is
closed in favour of this issue. Its analysis of the failure is still accurate and reusable.

See also the silent-close() defect, filed separately: it survives this fix independently.

[cdeust]

The silent-terminator defect referenced at the end is #135. It survives this fix independently: #134 makes close() resolve again, but the handler stays written so that any future failure to terminate the stream is invisible.

[cdeust]

One more thing to undo with the fix: the 3.1.0 CHANGELOG carries the false premise as published record.

### Removed
- The unused module-level `emit`, `close` and `reset` forwarders in
  `cortex_viz.server.graph_event_stream`. Nothing in this repository's history
  ever called them, ...

"Unused" and "nothing ever called them" are both untrue — graph_build_run called all three and graph_build_merge called emit. Worth correcting in the same change, otherwise the next person auditing this reads the removal as justified and the four broken call sites as unrelated.