feat(client): check correct state of h1 stream before sending request

[joelwurtz]

This PR replace #1176

A different approach is used here

Mainly we try to read 1 byte from the stream in a sync way, since we didn't send anything yet we should receive a WouldBlock io error, if we doesn't receive this it either means :

• the stream is in a different error so it should be closed
• the stream has some data left and then is in an incorrect state so we should close it
• the stream has been closed (so we should closed it also)

Retrying here should be safer since we don't consume request, i still use a middleware so user will opt in for this feature

I only handled the h1 connection closed error here, we could add other cases later when request could be retried safely in those cases

[fakeshadow]

#1358 should be fixing this.
During client pooling refactor a Ready trait is added before checking out a connection. It helps us despawn dead connection then in replace respawn. It handles h1 and h2 connection health without error reporting currently. Once the h3 support is also added we can explore how to better handling error case for it

[joelwurtz]

Indeed i was actually looking at the recent changes, will updates my pr accordingly, let me know if i can help in any way

[joelwurtz]

I'm closing this, it's no longer the case, thanks.