Core: Replay existing WAL/Parquet files on startup from foreign data dir

[decebal]

Problem

When allsource-core starts with DATA_DIR pointing at a directory containing existing Parquet files and WAL logs (written by an embedded Rust allsource-core client), it initializes fresh with 0 events instead of discovering and replaying the existing data.

docker run --rm --platform linux/amd64 \
  -v "/path/to/app/allsource:/data" \
  -e DATA_DIR=/data \
  -p 3900:3900 \
  ghcr.io/all-source-os/allsource-core:latest

Health endpoint shows:

{"total_events": 0, "tenant_events": 0}

Despite the data dir containing 400+ parquet files and an active WAL.

Expected behavior

Core should discover and replay existing Parquet + WAL data on startup, making all events queryable via the HTTP/WS API. This enables using Core as a read-only query layer over data written by embedded clients (e.g., Tauri desktop apps using allsource-core crate directly).

Context

Applications like Longhand embed allsource-core as a Rust crate and write events directly to WAL/Parquet. For debugging and MCP integration, we want to point the Core container at this data to query it — but Core doesn't pick up "foreign" data that wasn't written through its own bootstrap/tenant system.

Possible approaches

1. Auto-discover tenant data — scan storage/ for parquet files not associated with a system stream and create implicit tenant mappings
2. --import-dir flag — explicit CLI flag to replay a foreign data directory into Core's store
3. Read-only mode — CORE_READ_ONLY=true that skips bootstrap and just opens existing WAL + Parquet for queries