feat(weather): add official IMD district warnings and nowcasts

## What to build\n\nAdd an end-to-end India weather-alert slice from the official India Meteorological Department API reference at https://api.imd.gov.in/public/api_reference.html, starting with district warnings, district nowcasts, and station nowcasts.\n\nThe result must preserve issued-warning semantics. It must not convert ordinary forecast thresholds into alerts or present stale or unavailable IMD coverage as an all-clear state.\n\n## Acceptance criteria\n\n- [ ] The selected IMD endpoints, content types, access requirements, rate limits, update cadence, redirect behavior, response bounds, and stable identifiers are qualified before activation.\n- [ ] Redistribution, public-display, retention, and attribution rights are recorded. The integration remains disabled if those rights are not confirmed.\n- [ ] District warnings, district nowcasts, and station nowcasts normalize into the existing warning model without discarding issue time, validity, category, color or severity, message, district, state, and station provenance.\n- [ ] IMD-issued warnings remain distinguishable from derived forecasts and from NDMA SACHET relays.\n- [ ] Stable deduplication reconciles the same warning observed directly from IMD and through SACHET while retaining both provenance paths.\n- [ ] Source-specific failure or an empty response cannot erase other authorities or falsely report no warning.\n- [ ] The relevant dashboard surface provides India coverage, official attribution, original-source links, and an honest unavailable or stale state.\n- [ ] Health and seed metadata expose fetch freshness, content freshness, record count, coverage state, and last-known-good behavior.\n- [ ] Attribution, data-source documentation, generated catalog output, and focused endpoint, parsing, severity, deduplication, freshness, and degradation tests are included.\n\n## Blocked by\n\n- None — can start immediately.

Start with the official IMD API reference, then inspect the existing warning model, NDMA SACHET integration, dashboard surface, health and seed metadata, and generated catalog output. Define the endpoint and rights qualification before activation, then cover normalization, provenance, deduplication, freshness, degradation, attribution, and focused endpoint and parsing tests; done means honest India coverage without treating unavailable data as an all-clear.

Written by the indexing model from the issue text.