Record count estimation is too low

[sinui0]

A user has reported an error they received which comes from not preprocessing enough TLS record in MPC-TLS.

The relevant logic is here:

tlsn/crates/mpc-tls/src/config.rs

Lines 52 to 57 in d8747d4

	let max_sent_records = self
	.max_sent_records
	.unwrap_or_else(|| MIN_SENT_RECORDS + max_sent.div_ceil(16384));
	let max_recv_records = self
	.max_recv_records
	.unwrap_or_else(|| MIN_RECV_RECORDS + max_recv_online.div_ceil(16384));

Where we assume that the received records will be the maximum TLS record size (16kB). This turns out to not be a safe assumption.

We need to do 2 things:

1. Expose the configuration in tlsn_common::config::ProtocolConfig so that this can be overridden.
2. Come up with a better estimator.

[themighty1]

@sinui0, instead of estimating in advance, can the party allocate and execute the missing j0 blocks in the online phase?
Doing so will cost us some minor protocol latency at the expense of much better devex of not having to worry about such details.

[sinui0]

Im not sure we want/need to introduce that latency cost here once we have start proving the MAC key.

But aside from discussing that tradeoff, we're currently not able to dynamically allocate because of synchronization constraints between the MPC and ZK VMs. This will go away but will require work.

[themighty1]

closing since the remaining open issue is the only missing piece, we can move all conversation there.