Instead of using a global variable to check for a existing capture,
leverage the :org-roam plist property of the buffer-local variable
`org-roam-capture-current-plist`.
This prevents a lockout if the capture buffer is killed
This is a hotfix for #369. Currently, calling `org-roam-capture` sets
`org-roam-capture--in-process` to `t`, but cancelling the capture
process does not reset it to `nil`, causing false errors. The whole
nested org-capture processes workaround is brittle, this change moves it
as far back as possible, but the whole thing needs to be redesigned.
