feat(laqn): restore live ERG REST data source as LAQN-ERG

[ruaraidhd]

Problem

The default LAQN source fetches hourly data from the openair RData feed (londonair.org.uk/r_data/<SITE>_<YEAR>.RData). That's the right choice for bulk/backfill — it's ~600× faster than the old ERG REST path, which is why d2a0535 switched to it. But the RData files are regenerated only ~daily and their newest rows lag real time by a day or more.

That makes the default LAQN source unusable for live polling: a consumer requesting a short trailing window (e.g. "last 2 hours") gets nothing recent back. The switch to RData silently removed LAQN's live path, because the source has no separate notion of "latest" vs "archive" — there's one fetch_data, and it became the lagged one.

This bit a downstream live dashboard (Argus), whose 15-minute poller calls download("LAQN", …, last=<short window>) and, since the switch, has been getting only day-old data — so LAQN dropped out of the live view entirely. The upstream ERG API is healthy and current (hourly data within a few hours of real time); it just isn't wired to a data fetcher anymore (the default source uses it for metadata only).

Fix

Restore the ERG REST hourly-data path (recovered from d2a0535^) as a separate, non-primary LAQN-ERG source, mirroring the existing AURN-SOS-live / AURN-RData-backfill split:

• LAQN (unchanged) — openair RData. Stays primary; bulk/backfill keep their speed.
• LAQN-ERG (new, primary: False) — ERG REST Data/Site endpoint, current to within a few hours. Callers that need live data opt in explicitly.

Metadata is shared between both (the ERG JSON API). LAQN-ERG also exposes fetch_latest (last-4h max per site/measurand), matching the SOS sources.

Tests

13 new tests in tests/test_laqn.py:

• Fetcher — standard schema, source_network=LAQN, FINE→PM2.5, CO in mg/m3 / others ug/m3, UTC timestamps, empty-@Value rows dropped, unmapped species dropped, hits the ERG Data/Site endpoint, site codes upper-cased, no-data warning.
• Registration — LAQN-ERG registered as non-primary network source; default LAQN untouched.

Verification

Ran the deterministic blast-radius locally — 248 passed, 0 failed: test_conformance, test_registry, test_packaging, test_laqn, test_lmam, test_find_sites, test_api, test_sos, test_networks_portals_api (i.e. everything that exercises source registration, enumeration, conformance, and the registry reload-list cycles). The slow/integration (live-network, API-key) tests weren't run locally — that's for CI.

Follow-ups (not in this PR)

• Version bump + changelog entry at release time.
• Docs source-listing refresh to mention LAQN-ERG.
• Downstream: Argus sets aeolus_source = "LAQN-ERG" for live LAQN polling once a release carrying this ships (keeping LAQN/RData for backfill).
• This is a manual pre-figuring of the v0.5 network/backend split, where latest:/archive: backends and a FETCH_LATEST capability make "a network's live path" a first-class, testable thing — so this class of silent regression becomes a failing test at change time.

🤖 Generated with Claude Code