feat(bgp): optional attribute-255 clock on announcements

[digizeph]

Adds an optional per-prefix clock carried in attribute 255 (reserved for development, RFC 2042).

A [[prefixes]] entry with dev_attr255_interval_secs = N attaches a 13-byte payload to that prefix's announcements and re-announces it every N seconds. The field is optional, absent means no attribute, and 0 is rejected by config validation.

Payload layout: BGPKIT magic (6) + version (1) + refresh round (2, big-endian) + unix seconds (4, big-endian). The round is redundant with the timestamp once the interval is known, but it makes a single sample self-contained: a collector can count missed refreshes without knowing the cadence, and two samples reveal it. The container is deliberately not another originator's shape, so a code-255 sample is attributable from the payload alone.

The attribute is originated with OPTIONAL|TRANSITIVE explicitly, rather than Development's .into() default, which also sets PARTIAL. The first downstream speaker that does not recognize the attribute sets PARTIAL per RFC 4271, so collector copies of the same update can read e0 ff 0d .... Whether they do, and where the payload stops changing, is the measurement this exists for. Originating an unknown optional-transitive attribute is DFZ-visible by construction, so it stays scoped to our own prefixes and is detachable by re-announcing without it.

Runtime interaction: one refresh task per distinct interval re-dispatches the announcement and re-reads the effective prefix set on each tick, so a runtime suppression silences the clock and prefix add on a configured network keeps it. Clearing the field and reloading detaches the attribute; scheduling a newly added or changed interval takes effect after a daemon restart.

Tests cover the payload vector, the encoded bytes plus parse-back flags, the config roundtrip with zero rejection, and the socket-level refresh and runtime re-add paths.

Base is feat/collector-revision; retarget to main once that branch lands.

Note on CI: the Rust workflow only triggers for pull requests based on main, so this PR runs no CI. Local results on this commit: cargo fmt --check clean, cargo clippy --all-targets --all-features -- -D warnings clean, cargo test --all-features 49 passed / 0 failed (42 lib, 6 acceptance_peer, 1 archive_integration).