fix: re-use BufReader across poll_next invocations, instead of creating a new one

[kalvinnchau]

• re-use a single BufReader across poll_next calls, instead of creating a new one each time

  • this could cause buffered state to be lost across polls

• also create a re-useable Vec that we read into, and use std::mem::take once we have a full line to read off

  • std::mem::take will also clear the vector

this hopefully adresses some odd timing issues we saw when communicating with mcp servers

[salman1993]

cargo run -p mcp-client --example stdio_integration - failed for me

❯ cargo run -p mcp-client --example stdio_integration

...
2025-02-05T22:28:32.113621Z DEBUG mcp_client::transport::stdio: Stdout handler completed: ()
Error: McpServerError { method: "tools/list", server: "counter", source: ServerBoxError(Elapsed(())) }

[kalvinnchau]

cargo run -p mcp-client --example stdio_integration - failed for me

❯ cargo run -p mcp-client --example stdio_integration

...
2025-02-05T22:28:32.113621Z DEBUG mcp_client::transport::stdio: Stdout handler completed: ()
Error: McpServerError { method: "tools/list", server: "counter", source: ServerBoxError(Elapsed(())) }

hmmm can't replicate this error on my local machine or when running in a docker container

[baxen]

Looks good! Let's leave a note imo on what we think this is about, something like

// The buf reader will accumulate from stdin or similar stream 
// we clear one line at each iteration of poll_next from the buffer 

[baxen]

nit: if my understanding of the issue is correct, we don't need to keep this as a member and clear it below? seems simpler to instantiate a new one in each poll next instead. but let me know if i am missing something

[salman1993]

this works. i didn't pull the latest changes 🤦🏼