Instead of using a deque, store incoming messages in a native
buffered channel, if buffering is enabled.
In addition, modify the batching algorithm so the enforced delay
between consecutive `addMessages` invocations is applied after
all pending messages are processed. This acts as a "cooldown", rather
than a "warmup". This avoids the need for more complex timing logic to
dispatch batches, removes latency in adding messages when received
infrequently, and natively blocking the goroutine until messages are
received.
Because the message processing loop always performs a blocking read
first, it is appropriate for low-throughput environments just as much as
high-throughput ones.
The default value of batchSize has been changed to 10, with a zero
cooldown. This means that when messages are arriving faster than they
can be inserted into sqlite, they will automatically become batched in
groups of up to 10.
Use netip.Addr instead of storing addresses as strings. This requires
conversions at the database level and in tests, but is more memory
efficient otherwise, and facilitates the following.
Parse rate limit exemptions as netip.Prefix. This allows storing IP
ranges in the exemption list. Regular IP addresses (entered explicitly
or resolved from hostnames) are IPV4/32, denoting a range of one
address.
