Raise nem_data_model_start_time floor from 2009 to 2015

[nick-gorman]

Summary

Cherry-picked from #67 (commit 66189a8) — one-line change to src/nemosis/defaults.py raising the nem_data_model_start_time floor from 2009/07/01 to 2015/01/01, with an inline comment explaining why.

Implements the short-term fix recommended in #66.

Why

nem_data_model_start_time is only consumed by tables with search_type = "all" in processing_info_maps.py — i.e. cumulative-snapshot tables: PARTICIPANT, DUDETAIL, GENCONDATA, SPDREGIONCONSTRAINT, SPDCONNECTIONPOINTCONSTRAINT, SPDINTERCONNECTORCONSTRAINT, LOSSMODEL, LOSSFACTORMODEL, MNSP_INTERCONNECTOR, INTERCONNECTOR, INTERCONNECTORCONSTRAINT, MARKET_PRICE_THRESHOLDS (12 tables total).

For these, _set_up_dynamic_compilers sets start_search = nem_data_model_start_time, then _dynamic_data_fetch_loop iterates monthly from there to end_time. With the floor at 2009-07, the loop probes 72 months of pre-2015 URLs — all of which 404, because AEMO restructured the 2009-2014 archives into one zip per month containing all tables, vs the per-table monthly zips NEMOSIS expects from 2015 onward. As #66 notes: "Nemosis just throws an error, then moves on". Raising the floor skips that doomed scan window.

This is a speedup with no incremental data-loss impact — pre-2015 data already doesn't work on master today. The change just makes the failure faster (no minutes of 404 retries) and trims log noise.

What's not affected

• The 24-ish tables with search_type = "start_to_end" or "end" — those derive start_search from the user-supplied start_time/end_time directly, not from this default.
• Users who genuinely want to override the floor for testing — tests/end_to_end_table_tests/test_*.py already monkeypatch defaults.nem_data_model_start_time = "2021/05/01 00:00:00" to narrow scan windows; that path still works.

Future work

#66 keeps tracking the long-term aspiration of parsing the whole-of-month zip format AEMO uses for 2009-2014, which would restore access to those years. This PR doesn't close #66 — it implements the short-term mitigation Matt explicitly noted in that issue while leaving the deeper fix open.

No README change

README doesn't currently document a date floor anywhere, and this isn't a behavior change so much as a perf fix for an already-broken case. The inline source comment + #66 is sufficient context for future readers.

Test plan

• uv run pytest tests/ — 367 pass, 1 skipped, 3 warnings (all pre-existing, unrelated)
• CI green

Related: #66, supersedes the same change from #67.