Forwards the engine's log lines to Logger.

The Rust engine logs through the tracing facade and never installs a subscriber itself (docs/decisions/0009-observability-decisions.md, decision 3). When the :trellis_pg application starts, this process installs one in the Trellis NIF library, then drains what it queues into Logger, with domain: [:trellis] and the emitting Rust module as :target metadata. It needs no Trellis handle: every handle's lines come through it.

Configuration

# Forward events at this level and more severe ones. Defaults to
# `Logger.level()` when `:trellis_pg` starts. Rust's `trace` level
# arrives as `:debug`.
config :trellis_pg, log_level: :info

# Opt out: nothing is installed, this process doesn't start, and the
# engine's lines go nowhere.
config :trellis_pg, log_bridge: false

The level is read once, when :trellis_pg starts. The engine drops events above it before formatting them, so a later Logger.configure/1 can make Logger stricter but can't bring back a level the bridge filtered.

What can be lost

A log line can be dropped; logging never blocks or crashes the engine. The engine only appends to a bounded queue (10,000 lines), and this process drains it every 100 ms. If the queue fills, because this process is down, restarting, or falling behind, new lines are dropped and counted, and the next drain logs a warning with the count.

Composing your own subscriber

tracing's global subscriber lives inside the native library that links it, and the Trellis NIF links its own copy. So a subscriber another NIF installs never sees the engine's lines, and doesn't stop this bridge. Sending them elsewhere, for example through the crate's otlp layer, means building your own NIF around the trellis crate that installs that subscriber. If a subscriber in the Trellis NIF was installed first, this process logs a warning and doesn't start, and the engine's lines go to that subscriber instead.