fix(producer): handle drop_if_high_mem option

[thalesmg]

Fixes https://emqx.atlassian.net/browse/EMQX-13853

[zmstone]

maybe easier to understand and use if this is to control the max number of in-flight batches, instead of max inflight of calls ?
and by default 100 should be good.
like max_send_ahead for wolff_producer

[thalesmg]

it seems to already be the number of inflight messages, despite the name

[zmstone]

for Kafka, emqx uses default value max_inflight = 10 in the schema

[thalesmg]

... as it counts individual messages, not calls:

pulsar-client-erl/src/pulsar_producer.erl

Lines 825 to 830 in 1865307

	FinalBatchSize = length(FinalBatch),
	send_batch_payload(FinalBatch, SequenceId, State0),
	Requests = Requests0#{SequenceId => ?INFLIGHT_REQ(QAckRef, FromsToMessages, FinalBatchSize)},
	InflightCalls = InflightCalls0 + FinalBatchSize,
	pulsar_metrics:inflight_set(State1, InflightCalls),
	State2 = State1#{requests := Requests, inflight_calls := InflightCalls},

[thalesmg]

and I tested with 10, and it lead to a lot of queuing, hurting throughput

[thalesmg]

... though, I just tested again, and it worked fine with 10 🤔

[zmstone]

this will cause caller to exit if caller is not trapping exit. but it has to happen, it's the remote-shell process (when trying to manually stop).