fix: retry startup auto-run while the database worker is unavailable

[heryr]

Problem

With auto: true the startup migration run was attempted exactly once. When miiigrate registered before the database worker — the common case under docker compose or an engine-managed boot, where registration order is not guaranteed — the run failed with DATABASE_WORKER_UNAVAILABLE and was never retried, silently leaving migrations unapplied until a manual migrate::up.

More generally, a normal cold boot produced apparent errors: the configuration::register/get calls gave up after ~16 s and crashed the worker (leaving the restart policy to recover), and iii-helpers' OTel side channel logged every pre-connection attempt at ERROR.

Fix

• Auto-run retry — the startup run retries on DATABASE_WORKER_UNAVAILABLE with capped exponential backoff (1 s doubling up to 10 s, 2-minute budget). Any other error — and budget exhaustion — remains loud and terminal for the auto run only: the worker stays up so migrate::status and migrate::up are available for diagnosis and manual recovery, unchanged from before.
• Configuration RPC retry — transient failures (timeout, not connected, function not registered yet) are retried with capped backoff under the same 2-minute budget instead of crashing the worker; real errors from the configuration worker still fail fast.
• No duplicate ERROR noise — the default tracing filter silences iii_helpers::observability::telemetry::connection, which logs every pre-connection attempt at ERROR (hardcoded in the helper) while iii-sdk already reports the identical condition as a WARN retry. Setting RUST_LOG replaces the filter entirely.

Docs updated: docs/configuration.md (auto row) and docs/changelog.md (Unreleased).

Validation

• cargo build, cargo test (47 tests), cargo clippy — clean.
• Reproduced end to end in a dockerized stack (engine 0.21.6 + Postgres 16 + database worker, miiigrate built from this branch), cold start with no ordering guarantee. Zero ERROR lines; the full boot log is INFO plus three legitimate transient WARNs:

INFO  miiigrate: starting name="miiigrate" …
WARN  iii_sdk::iii: failed to connect; retrying error=IO error: Connection refused (os error 111)
INFO  iii_sdk::iii: iii connected address=ws://iii:49134
INFO  miiigrate: configuration loaded db=primary dir=./migrations auto=true
WARN  miiigrate: database worker not ready; retrying auto migration run … retry_in_s=1
WARN  miiigrate: database worker not ready; retrying auto migration run … retry_in_s=2
INFO  miiigrate::handlers::up: applied migration=20260101120000_create_users.sql
INFO  miiigrate: auto migration run complete …
INFO  miiigrate: miiigrate worker registered 4 functions, waiting for invocations