serial console error message on node that is not running is unhelpful

[rcgoodfellow]

If an instance is not running for some reason and an attempt to get a serial console is made, we get the following.

$ ./solo serial turbolift
Entering VM console.
Use ^q to exit.
Press enter to continue.
Error: Anyhow(failed to create serial websocket stream

Caused by:
    HTTP error: 200 OK)

Which is not at all helpful.

This can be reproduced as follows with a simple one node topology.

use libfalcon::{cli::run, error::Error, unit::gb, Runner};

#[tokio::main]
async fn main() -> Result<(), Error> {
    let mut d = Runner::new("solo");

    // nodes
    let turbolift = d.node("turbolift", "helios-2.9", 8, gb(8));
    run(&mut d).await?;

    Ok(())
}

pfexec ./solo launch
# wait for topology to come up
pfexec pkill propolis-server
./solo serial turbolift

[iximeow]

the HTTP error: 200 OK here, at least, most immediately seems to come from the websocket upgrade as part of InstanceSerialConsoleHelper::new getting an.. HTTP 200 back, instead of the 101 Switching Protocols that one would normally expect.

I don't see how propolis-server would do that given the impl at https://github.com/oxidecomputer/propolis/blob/cae89ab26cf66244fa3cd6796d17ce5c5049c928/bin/propolis-server/src/lib/server.rs#L414-L430, but somewhere between dropshot and this function it seems like it must be possible..