MetaDataStore reloads on Brightway on_database_write (hold under SafeBWConnection)

A full Database.write replaces inventory without per-activity signals, so the Metadata store refreshes by listening to Brightway’s on_database_write—not via call-site helpers on importers. That blinker can fire while a worker still holds SQLite (SafeBWConnection open), and modal processEvents can run a GUI reload too early (NULL names → UI nan). Decision: request_metadata_reload(name) schedules immediately, or while hold_metadata_reloads() is active only remembers names. On SafeBWConnection exit: close peewee, then release_metadata_reloads() (schedules each remembered name). Do not wrap Brightway’s send. Parameterized-flow rebuild keeps normal blinker timing via ABSignals (ADR-0010). Writers must emit (signal=True or equivalent .send()); silent writes stay stale. Register→add_database may load early; the write notification is authoritative. Excel/BW2Package must not schedule metadata reload themselves.

Considered Options

• Call-site post-thread schedule_* on importers — two paths only; re-teaches every writer.
• Wrap/defer Brightway on_database_write.send — correct but hard to reason about.
• Eager schedule on blinker with no hold — races worker + processEvents.

Consequences

• Only Metadata store reload scheduling is held; other blinker listeners see the write when Brightway sends it.
• Reload scheduler is created on the GUI thread at loader init; schedule_database_metadata_reload only emits.
• Holding is per-thread (names is None vs a set); SafeBWConnection is only used once around ABThread.run_safely.