fix(fetcher): release fetched payload after each worker iteration

[Kyzgor]

Fixes Issue

Closes #844
Closes #770

Changes proposed

Fixes a memory high-water mark in the data fetching engine: a fetch worker held its last
fetched dataset (and HTTP response) in memory while it was idle.

• opal_common/fetcher/engine/fetch_worker.py: null all five per-iteration locals
  (event, callback, fetcher, res, data) in the worker's finally block, which runs
  after the exception handlers so they still see event, and pre-initialize
  fetcher = res = data = None before the try (those three may be unassigned if an early
  exception occurs). An idle worker now holds no payload between queue items.
• opal_common/fetcher/tests/fetch_worker_retention_test.py (new): a regression test that an
  idle worker does not retain its last fetched payload. It uses a weakref, so it fails if the
  cleanup is removed.

Root cause

fetch_worker runs while True: event, callback = await queue.get(); .... In CPython a frame
keeps its locals bound until they are reassigned or the frame exits, so a worker blocked on the
next queue.get() still referenced the previous iteration's event/fetcher/res/data,
which pins the last fetched dataset and its aiohttp response. With FETCHING_WORKER_COUNT
workers (default 6) this retained up to N full datasets, proportional to payload size, released
only on process restart. OPA's own memory was unaffected (the data is correctly stored there);
the retention was entirely client-side in the worker frames. That is also why it looked like a
leak yet heap inspection after things settle found nothing: the objects free as soon as a worker
handles a new item, but idle workers keep the last one.

This is the same mechanism behind both reports. #844 triggers a fetch on every reconnect
("Initial load"), and #770 triggers one on every periodic datasource refresh.

Check List (Check all the applicable boxes)

• I sign off on contributing this submission to open-source
• My code follows the code style of this project.
• My change requires changes to the documentation.
• I have updated the documentation accordingly.
• All new and existing tests passed.
• This PR does not contain plagiarized content.
• The title of my pull request is a short description of the requested changes.

Screenshots

N/A. The behaviour is memory usage, quantified below.

Note to reviewers

Before/after, from a local harness driving the real DataUpdater -> fetch -> store path (full
reproduction steps available on request):

Scenario	Before fix	After fix
#844 reconnect loop (2 MB payload, 12 reconnects, default 6 workers)	staircase +~73 MB, ~6x payload retained per reconnect, never released	plateau, +~5 MB total; retained dataset copies 6 -> 1
#770 periodic refresh (13 MB payload, real OPA, 12 cycles)	climbs to ~630 MB high-water mark (about 6 workers x ~80 MB), held until restart	flat ~159 MB (+0.11 MB/cycle, i.e. noise)

The defining signature of the bug was that the retained-dataset count equalled the worker count.
After the fix the retained count is 1 (measured with the default FETCHING_WORKER_COUNT of 6),
and since the retention was structurally one dataset per worker it no longer scales with the
pool size. malloc_trim does not reclaim the pre-fix growth, which confirms genuine retention
rather than allocator fragmentation.

Correctness: the null-out is in finally, which runs after the except blocks that call
engine._on_failure(err, event), so the failure handler still receives the real event. The
existing fetcher and data_updater tests pass (10), and with the new regression test all 11 pass.

Scope: this PR only changes the worker frame-retention behaviour. During the investigation I also
found a small, separate unbounded object accumulation in the pub/sub reconnect path
(asyncio.as_completed in fastapi-websocket-pubsub). It grows one task and one queue per
reconnect but is only KB-scale and lives in a different repository, so I've raised it separately
and it is not part of this change. The growth fixed here is a bounded high-water mark under
normal (gc-enabled) runtime.

CI note: the security/snyk check tends to error on fork PRs (org token/quota) rather than
indicating a code problem. Happy to have a maintainer re-run it.

[netlify]

✅ Deploy Preview for opal-docs canceled.

Name	Link
🔨 Latest commit	a93b6a0
🔍 Latest deploy log	https://app.netlify.com/projects/opal-docs/deploys/6a2706f7de4a660007b728dd