Sliding window counter rate limiter.

A fixed window counter lets a client send limit requests at 12:00:59 and limit more at 12:01:00: twice the limit in one second. The sliding window log fixes that but stores every timestamp. Day 3, section 2.3 gives the production sweet spot: keep only the count for the current fixed window and the previous one, and weight the previous count by how much of it still overlaps the sliding window.

window      = int(now // window_seconds)        # which fixed window we're in
position    = now % window_seconds              # seconds into it
prev_weight = (window_seconds - position) / window_seconds
estimate    = prev_count * prev_weight + curr_count

Allow the request when estimate < limit, and only then add 1 to curr_count.

Implement SlidingWindowCounterLimiter(limit, window_seconds, clock):

Rules the tests check:

• When time moves into the next window, the current count becomes the previous count and the current count starts at 0.
• When time skips more than one window, the previous count is 0: nothing from two windows ago overlaps.
• Rejected requests are not counted.
• Store two counters per key, not timestamps.
• Read time only through clock.now().

The lesson's worked example (limit 10 per 60 s): at 12:01:15 the previous minute had 8 requests and this minute has 3. The previous window still overlaps 45 of the last 60 seconds, so 8 × 0.75 + 3 = 9 < 10 and the request is allowed.