fix(liquidity): keep activation debounce across drain chunks

[TaprootFreak]

Problem

verifyRule cleared the rule's activation-debounce timer (ruleActivations) unconditionally right before executeRule. After each pipeline completes and the rule returns to Active, the next cron cycle re-arms the timer and must wait the full lmActivationDelay again — before the next chunk. For a redundancy rule whose orders are liquidity-capped (best-price chunk, e.g. an exchange sell), this serialized the drain to one chunk per delay (~15 min), turning a large offload into a multi-hour trickle.

Fix

• Remove the unconditional ruleActivations.delete before executeRule. The timer is armed once when the condition first appears and cleared only when the condition resolves (the else branch). Follow-up chunks of an ongoing drain then fire at the normal every-minute cadence.
• Clear the timer at the actual pause transition: handlePipelineFail calls a new resetActivation(ruleId) on the liquidity service right after rule.pause(), so a rule paused by a failed pipeline re-debounces on its next activation — regardless of whether reactivation happens via the cron or the manual admin endpoint. (.pause() has exactly one call site, so this covers every pause.)

Behaviour is unchanged for rules that satisfy their need in a single pipeline and for deficit paths. Adds regression tests for the debounce invariant and the pause→reset wiring. Verified locally: tsc, eslint, jest.

[TaprootFreak]

Three review passes to reach zero findings. Pass 1 flagged missing regression coverage for the debounce invariant (added). Pass 2 found that clearing the timer inside verifyRule had a race with the manual reactivation endpoint; the reset was moved to the actual pause transition (handlePipelineFail), which causally precedes any reactivation. Pass 3: zero findings on both conformance and logic, verified against the module's tests.