Fix the window reset in a tenant rate limiter

Your multi-tenant API limits each tenant to a number of requests per time window. The first version reset each tenant's count with a timer scheduled on every request, so a burst set off a cascade of resets that wiped counts in the middle of a window. The new version takes the tenant from the authenticated caller and works out the window from each request's timestamp.

An assistant wrote rate_limit_decisions(tenant_ids, timestamps_ms, max_per_window, window_ms). It is in the editor. It has a bug. Find it and fix it.

Input and output:

• tenant_ids: the authenticated tenant for each request, in arrival order. Empty means the caller is not authenticated.
• timestamps_ms: each request's time in milliseconds. They never decrease.
• max_per_window: how many requests a tenant may make in one window.
• window_ms: the window length in milliseconds.
• Return one boolean per request: true if it is allowed.

Rules the tests check:

• If the two lists differ in length, use only the first n requests, where n is the shorter length.
• A request with an empty tenant is denied and counts toward nothing.
• If window_ms or max_per_window is 0 or less, deny every request.
• A request's window is its timestamp divided by window_ms, rounded down. With window_ms 1000, 0 to 999 is one window and 1000 to 1999 the next.
• Each tenant has one count, for the window of its latest counted request. When a tenant's request falls in a different window from that one, its count starts again from 0.
• Every counted request adds 1 to its tenant's count, whether it is allowed or not. It is allowed when the count is then at most max_per_window.