feat(cliprdr): add a clipboard sync loop detector

[Greg Lamberson (glamberson)]

Summary

• An embedder bridging CLIPRDR to a local OS clipboard commonly hits a
  feedback loop: content copied on one side syncs to the other, the
  other side's own change notification fires for that same content, and
  the embedder syncs it back, forever. ironrdp-cliprdr has no visibility
  into the local OS clipboard on either side by design (that lives
  entirely in the embedder's CliprdrBackend implementation), so it
  can't detect this on its own, but nothing in the crate gives an
  embedder a building block to break the cycle either. Every consumer
  bridging to a real desktop clipboard ends up needing to write the
  same hash-and-time-window correlation logic themselves.
• Adds LoopDetector: hashes recent format lists and content by source
  (Remote/Local) within a configurable time window, so an embedder can
  ask "would syncing this out right now just echo what the other side
  just sent" before acting, plus an optional per-source rate limit as
  a belt-and-suspenders guard against update storms.
• Every method that needs the time takes an explicit now_ms rather than
  reading a clock itself, matching CliprdrBackend::now_ms()/elapsed_ms()
  already on this crate for the same wasm/test-determinism reasons.
• Hashing uses DefaultHasher rather than adding a crypto dependency:
  nothing here defends against an adversary, it only needs to avoid
  mistaking two different clipboard payloads for the same one within
  one process's own recent history.
• would_cause_loop takes an explicit source parameter, matching its
  siblings would_cause_content_loop/should_skip_sync. An earlier
  version of this algorithm I'd written elsewhere hardcoded that
  direction on would_cause_loop specifically, which was an
  inconsistency with its own siblings rather than an intentional
  asymmetry; fixed here.

Validation

cargo xtask check fmt/lints/tests/typos/locks all pass. Tests live in
ironrdp-testsuite-core since this crate sets [lib] test = false.

Notes

No public API break: this is a new module with entirely new public
types, no existing signature changes.