Separate add_rx from Bus

This might be related to or a duplicate of #19 , but it's not clear based on the wording of the original issue.

The problem with `add_rx` being tied to `Bus` occurs in my use case where I want to:

1. Send the `Bus` to a separate "dispatch" thread.
2. Dynamically add/remove receivers over time on the main thread.

Once the `Bus` has been moved to the dispatch thread, it can't be used on the main thread to create more receivers, thus fixing the number that I can create. It is technically possible to send messages to create more and send them back via channels, but I have another idea I'd like to pursue to see if it's any better.

What I propose is creating a secondary interface (let's call it `ReadHandle` for now), which could be created by a method on `Bus` and would also implement `add_rx`.

Internally, it would only involve taking clones of `Bus.state` and the senders `Bus.leaving.0`, `Bus.waiting.0` when constructed, so it would be mostly zero-cost. The one major change would be making `Bus.readers` atomic and moving it to `BusInner`, but we can at least try it and see how it affects benchmarks.

Start by reading the existing Bus and add_rx implementation, then inspect how Bus.state, Bus.leaving.0, Bus.waiting.0, and Bus.readers are used. Compare the proposal with issue #19 and review the benchmark coverage before deciding on the ReadHandle design. Done means receivers can be added or removed from the main thread after Bus moves to a dispatch thread, with synchronization behavior and benchmark effects addressed.

Written by the indexing model from the issue text.