Optimize connection use in stream_conversations

[insipx]

stream_conversations with callback pulls a new connection from the database multiple times, despite theoretically having the ability to do one of
1.) hold on to one connection for the duration of the stream
2.) pull one connection for all its operations before adding it back to the pool

Here's where we pull connections:

• transaction_async pulls a new connection for processing a new welcome
• pulls another connection again after the transaction for other operations
• DM groups will pull yet another connection when checking for DM groups.

instead, we can create a new transaction function wrapping the existing one transaction_async_with_provider to reuse a provider, and then reuse that connection for the subsequent operations. this should reduce some churn in the Pool upon getting a new welcome/group in the stream

Then require that process_streamed_welcome accept a provider

This will allow us to instantiate a provider at the head of the stream message and use it for all ops

[insipx]

done in #1390