Liveness probing — Boundary stage of the resilience CPPB pipeline.

Single responsibility: ask the underlying milvus client whether it can
actually reach the server, with a tiny TTL cache so a burst of probes
amortises to one RPC.

Transport-agnostic: dispatches through `milvus-clj.client/ILivenessProbe`
(extended by every transport's defrecord). Adding a new transport
requires zero changes here — that is the point of the protocol seam.

Why a separate ns from reconnect/retry: each of probe / reconnect /
retry has its own rate of change. The probe surface is the most stable
(one cached read); reconnect-loop algorithm and retry budget knobs
change more often. Splitting them keeps each ns small and individually
testable (mock the protocol; no need to mock a loop).

Time discipline: never call `System/currentTimeMillis` directly — go
through `hive-ttracking.clock/now-millis` so tests can pin time.

Reconnect-loop ownership — Boundary stage of the resilience CPPB
pipeline.

Single responsibility: drive the singleton milvus client back to a
reachable state after a transient failure. The loop's success
criterion is an actual probe round-trip, NOT the post-`connect!`
atom state — that distinction is the bug fix vs the prior design
where `connect!` always 'succeeded' (it just sets the atom) and the
loop exited even when the network was still down.

Decoupled from probe.clj (which owns the cache) and retry.clj
(which owns the reactive pipeline) per SRP. This ns owns ONE atom
(`reconnect-state`) and its loop future.

Time discipline: never call `System/currentTimeMillis` directly — go
through `hive-ttracking.clock/now-millis` so tests can pin time.

Reactive try / classify / retry pipeline — Pipeline stage of the
resilience CPPB pipeline.

Single responsibility: wrap one RPC body in a fault-tolerant
pipeline that classifies failure, kicks the reconnect loop, awaits
verified recovery, and retries exactly once. Public surface:

  - `with-auto-reconnect` — function form, takes a thunk
  - `resilient` — macro sugar that combines `ensure-live!` with
    `with-auto-reconnect` for use inside protocol method bodies

Composes Promote (`failure/classify`, `circuit/check`) over Boundary
(`probe`, `reconnect`). Stays pure with respect to the singleton
client — never mutates `default-client` directly; only orchestrates
the Boundary nses that do.

Migration note: this ns replaces `hive-milvus.store.health`. The old
ns kept circuit, probe, reconnect, classify, retry all in one file
(5 responsibilities). This split means each concern can evolve
independently.

Proactive periodic liveness scheduler — Boundary stage of the resilience
CPPB pipeline.

Single responsibility: own a timer that fires a health check on a fixed
cadence, so a connection that dies while the store is IDLE (no traffic to
trigger the reactive `retry` path or the on-demand `ensure-live!` gate)
still heals. This closes the gap where auto-heal was 100% traffic-driven:
after a network blip with no in-flight RPC, nothing re-probed and the dead
client sat dead until the next operation happened to touch it.

Why this ns owns ONLY the timer (SRP + DRY):
- The health *decision* — 'is it dead, and if so start a heal loop WITHOUT
  disturbing one already in flight' — already lives in `retry/ensure-live!`
  (reads cached liveness, kicks the reconnect loop only `when-not` it is
  already running). Re-implementing it here would duplicate the guard and
  risk calling `reconnect/kick!` mid-heal, which drops the fresh client on
  every tick. So the tick DELEGATES the policy and keeps only the
  scheduling (Boundary) concern.
- Complements, does not replace: reactive retry (on failure) + preemptive
  ensure-live! (at op start) stay as-is; this adds the idle heartbeat that
  neither covered.

Lifecycle: started from `store.lifecycle/connect!` once the config-atom
exists; stopped from `store.lifecycle/disconnect!` before teardown, so the
timer can't kick a heal loop right after the operator closed the store.

Time discipline: never call `System/currentTimeMillis` directly — go
through `hive-ttracking.clock/now-millis` (mirrors probe/reconnect).