perf(server): count active tasks in SQL before create

[kvnloo]

Problem

Every createTask call currently loads the entire durable task table and filters terminal rows in JavaScript to enforce the 100-active-task cap. Terminal tasks are retained, so this check gets more expensive over the lifetime of a workspace.

A downstream PGlite benchmark with 2,000 ~2KB task rows measured roughly:

• full list()+filter: ~282.7ms / ~3.0MB transferred
• SQL count: ~54.3ms / ~50B returned

Change

Add Store.countActiveTasks() and push the terminal-status exclusion into SQL. COALESCE preserves the previous behavior where a status-less task counts as active.

createTask still applies the same >=100 cap and the same terminal set.

Regression

The persistence test compares the SQL result to the previous list()+filter semantics across all task statuses, including a status-less record, and verifies owner/kind isolation.

No cap value, task lifecycle, or API behavior changes.