[milutinke]

Bump DMCBK to 0.1.0-preview.8, update UMPK to 0.9.0-beta.7, and cut repeated allocations in Beacon timer polling. Reuse dispatch plans until registrations change, and create firing records only when a timer is due. Reloads and removals invalidate the cache; each active tick keeps its snapshot across awaits.

Mean of five alternating process pairs, before → after. Both variants use UMPK beta.7; measurements precede the DMCBK package version bump.

Workload	Scripts	Allocation B/poll	CPU µs/poll	Process RAM MB
Idle	1	1,864 → 352	1.39 → 0.50	108.09 → 107.04
Idle	32	18,384 → 352	10.85 → 2.89	109.34 → 94.08
Idle	128	70,705 → 353	33.66 → 8.00	111.58 → 82.38
Mixed	1	6,717 → 5,211	8.76 → 7.74	106.46 → 106.70
Mixed	32	106,957 → 89,083	141.55 → 123.03	113.76 → 112.85
Mixed	128	419,231 → 349,411	320.79 → 275.69	125.98 → 123.87

These are unpaced, headless workloads. A poll checks recurring and one-shot timers. Mixed activity also advances virtual time 50 ms and dispatches a seeded chat, inventory or player-list event. RAM includes the runtime and dependencies; live Minecraft capacity was not measured.

Validation: 1,963 tests passed, including timing, ordering, reload/removal and collectibility regressions. Release build and formatting passed. All nine packages packed, and all affected samples, including WebBackend, built against the exact local packages with UMPK beta.7.