fix(release): repair make publish crate set, order and dev-dep cycles

[lxsaah]

make publish could not complete. It failed at cargo publish -p aimdb-tokio-adapter, and no reordering of the list could fix it: three
workspace library crates were absent from the list entirely, and the
dependency graph was cyclic once dev-dependencies were counted.

Three changes, no code movement.

Add the three missing connectors. aimdb-uds-connector,
aimdb-serial-connector and aimdb-tcp-connector are published API —
aimdb-client already exposes all three behind its transport-*
features, and cargo publish requires every dependency to resolve on the
registry, optional ones included.

Replace the hand-numbered 16-step list with a PUBLISH_ORDER variable
driving a loop, so the order has one source of truth and the step
numbering derives from it. The order is topological over normal and build
dependencies, verified against cargo metadata. PUBLISH_NO_VERIFY
carries the two crates that skip verification, keeping the existing
behaviour explicit rather than buried in one line of the sequence.

Drop version = from the workspace-internal dev-dependencies, keeping
path. Cargo omits path-only dev-dependencies when packaging, so the
cycle disappears at publish time while cargo test still resolves
locally. Five needed it, not the three first identified: the edge
aimdb-tokio-adapter -> aimdb-client appears in four of the five
elementary cycles, and aimdb-embassy-adapter -> aimdb-core = "1.1.0"
would have built the packaged tests against the previously published core
rather than the code being released. Conversely,
aimdb-websocket-connector -> aimdb-tokio-adapter forms no cycle —
nothing reaches the websocket connector from the tokio adapter — but it
carries the same stale-version hazard and is fixed on that basis.

Also mark remote-access-demo as publish = false. It was the only
example crate missing the flag, which left it as a 20th publishable crate
that no publish list mentioned.

Verified: the target runs end to end against a stubbed cargo, emitting
19 publishes in the listed order with --no-verify on exactly the two
intended crates, and aborting on the first failure; cargo metadata
reports no versioned workspace-internal dev-dependencies remain and the
list matches the publishable set exactly; the packaged-manifest behaviour
was reproduced on a scratch workspace, whose .crate carries an empty
[dev-dependencies] section; and cargo check --all-targets passes for
every affected crate.

make publish is still not runnable end to end: aimdb-embassy-adapter
needs embassy-sync's poll_next_message / poll_changed, which are
merged upstream but absent from the released 0.8.0 and supplied locally
through [patch.crates-io]. That is recorded on PUBLISH_NO_VERIFY and
is the remaining blocker.

Refs #242

Co-Authored-By: Claude Opus 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01GgvcFC4S8Kii5H6xzW56mi