- history/worker: fix deadlock by moving stale job requeue out of claimNextJob
into dedicated staleJobRequeuer goroutine (runs every 2 min)
- history/service,tx_apply,cross_entity: add deferred=true mode — write events+
snapshots but skip projection updates; queue recompute after commit
- ingest/service: IngestHardwareDeferred uses deferred mode; CSV workers up to 8
(INGEST_CSV_WORKERS env); serial/prefetch lookups use normalize.SerialKey
- api/ingest: JSON /ingest/hardware now async (202 + job_id); new GET
/ingest/hardware/jobs/{id} endpoint; CSV already async
- history/admin_cancel: replace per-event softDelete loop with batchSoftDeleteEvents
using IN-clause chunks of 500 to prevent request timeout on large deletes
- normalize: new internal/normalize package with VendorKey, VendorDisplay,
VendorDisplayPtr, SerialKey, FirmwareKey
- ingest/parser_hardware: vendor fields use normalize.VendorDisplayPtr
- migrations/0021_ci_identifiers: change identifier columns to utf8mb4_unicode_ci
(case-insensitive) in parts, machines, observations, machine_firmware_states
- bible submodule: update to add identifier-normalization contract
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fixes 400 Bad Request error when ingesting hardware snapshots that include
PCIe link speed information (link_width, link_speed, max_link_width, max_link_speed).
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
