fix: same-timestamp events dropped, silent timestamp fallback, stale docs ref, Cargo cruft

[Donemmanuelo]

AuditBridge version or image digest

main @ 704a08b (unreleased, Cargo.toml 0.1.0)

Steps to reproduce

1. Two audit events share the same RFC3339 timestamp (NetBird returns them in
   one poll; both are delivered and the per-sink cursor advances to that
   timestamp). A third event with the same timestamp arrives in a later poll.
   process_cycle filters with strict ts > cursor, so the third event is
   dropped permanently — silent event loss.
2. An event with an unparseable timestamp hits the Loki encoding:
   timestamp_to_nanoseconds silently substitutes Utc::now(), rewriting the
   audit time to "right now", non-deterministically across polls.
3. src/netbird.rs comments reference the README's "Scaling Limits" section,
   which does not exist.
4. Cargo.toml carries commented-out dependencies (tokio-rusqlite,
   rusqlite, dotenvy).

Expected behavior

Events at the watermark boundary are delivered at least once (boundary event
IDs tracked per sink alongside the timestamp). Unparseable timestamps produce a
deterministic fallback and a logged warning instead of silently rewriting to
now(). Code comments match the README; Cargo.toml is clean.

Actual behavior and sanitized logs

1. pending.retain(|e| ts > last_ts) excludes same-timestamp events forever.
2. unwrap_or_else(|_| Utc::now()...) substitutes the current time silently.
3. Stale reference to a section that was never written.
4. Dead commented-out dependency lines.