Pinning is critical operation whose failure will soon lead to client
being killed with SIGBUS. WCFS correctness also depend fundamentally on
pin operation, if started, to be handled by the client.

-> rework the READ handler not to cancel pin if a READ interrupt comes
   in from the OS client.

Do this via organizing WatchLink.serveCtx and running pins under this
context instead of under READ context. Later we will adjust pins to also
cancel this context on any error.

Test is, hopefully, TODO.

/reviewed-by @levin.zimmermann
/reviewed-on nexedi/wendelin.core!18