[lxsaah]

Description

Stage 1 of 2 for #390, into the #392 feature branch.

• ConnectorHandle { builder: u32, index: u16 } in aimdb-core::connector, exported at the crate root. It's Copy, Eq and Hash, and holds no string. It's named ConnectorHandle because the crate root already exports the session-layer Connection.
• AimDbBuilder::connect(&mut self, c) -> ConnectorHandle registers the connector exactly like with_connector, and the index saturates at u16::MAX.
• Each builder gets an id from a static AtomicU32 starting at 1, so tests can use 0 as a fixed id. Every target in rust-toolchain.toml has this atomic, so no portable-atomic is needed.
• with_connector's docs point to connect.

Unlike the plan in #390, RecordRegistrar gets its builder_id field in stage 2, together with OutboundLink::on / InboundLink::on and the check that reads it. In this stage nothing would read it.

Tests

• builder::connect_tests::connect_hands_out_one_handle_per_connector: indices 0 and 1, the same id within one builder, a different id in another builder, and connectors kept in order.
• typed_api::tests::connected_connector_serves_links: a connector added with connect serves build() and inbound routing like one added with with_connector.

Checks

CI doesn't run on PRs into the feature branch, so I ran these locally:

• aimdb-core tests with std,tracing,observability (307) and with alloc
• the thumbv8m build
• all nine of core's clippy lines from the Makefile (-D warnings), including thumbv7em with defmt
• cargo doc with -D warnings

Related Issue

• Part of [FEAT] AimDbBuilder::connect returns a ConnectorHandle; links through a handle need no scheme #390 / [FEAT] reg.input / reg.output with connection handles, so a wrong wiring is a compile error #388 (PR feat: reg.input / reg.output, connector handles and publish / subscribe (#388) #392)

🤖 Generated with Claude Code