Download the airport database instead of building it

[RISCfuture]

This is what the rest of the work was for. The app asks for the store
that was built for this cycle, expands it into a new generation, and
switches to it — so the minutes it spent inserting a decoded property
list into SwiftData become a transfer and a decompression.

Cycles are asked for by name and walked backwards, one at a time. A
manifest states the window it is effective for, which a mutable "latest"
pointer could not, and a cycle whose publish failed therefore costs the
pilot the previous cycle's data rather than four weeks with no update at
all.

The published schema fingerprint is checked before anything is
downloaded, let alone opened. SwiftData answers a near-miss by migrating
the store rather than refusing it, so a build that opened first would get
a silent, slow migration where a clean fall back to importing was wanted.
The download is checked against the digest its manifest published, and
the expanded store still has to open and hold airports before anything
switches to it.

Every step may fail and none of them is fatal: a cycle nobody published,
a store built for a different schema, a truncated transfer — each falls
back to importing the property list, which is still published and still
works. That is what keeps this an optimization rather than a dependency,
and it is why the import path was kept rather than replaced.

Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01471376Uhug2TYZN8RDFGtq

---

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}

🤖 Generated with Claude Code

https://claude.ai/code/session_01471376Uhug2TYZN8RDFGtq