Describe the feature

the way the rpc cache works is a single loop over actions:

disk lookups also reuse that single channel and send actions back to that sender:

in the meantime the oneshot response senders wait for that action:

these response channels remain in the lru until the action arrives in the mainloop.

This is problematic if the spawned task panics because then nothing is emitted back to the mainloop.
for example if this operation panics

we can solve this by introducing a drop aware sender wrapper that sends an action on drop, if not consumed:

like:

struct ActionSender {
  blockhash: B256,
  tx: Option<UnboundedSender<...>
}

impl ActionSender {
  fn send_block(self,...)
}

impl Drop for ActionSender {
  if let Some(tx)
   ... send error
}

Additional context

No response