Latency Percentile Tracker

You are building a latency monitor for a streaming service. Process the list operations in order. Each operation is one of:

• ["record", latency]: store one latency sample (a non-negative integer, in milliseconds). Returns nothing.
• ["percentile", p]: report the nearest-rank p-th percentile of every sample recorded so far.

Nearest-rank percentile. Let n be the number of samples recorded so far and imagine them sorted in ascending order (duplicates kept). The rank is r = ceil(p * n / 100), and the answer is the r-th smallest sample (1-based). For example, with samples [10, 20, 30, 40], the 50th percentile has rank ceil(2) = 2 and is 20; the 51st percentile has rank ceil(2.04) = 3 and is 30. If no samples have been recorded yet, the answer is -1.

Return a list with one answer per "percentile" operation, in the order those operations appear.

Samples and queries are interleaved, so a query only sees the samples recorded before it. Aim for something better than re-sorting all samples on every query.