### Motivation & Modifications

Add a script to download and stage artifacts so that we can just use a
`svn add *` command to add them to the central repo.

### Verification

Take the 3.0.0-candidate-3 release as example, after configuring the
`GITHUB_TOKEN` environment variable, you only need to run:

```bash
# See https://github.com/apache/pulsar-client-python/actions/runs/3709463737
$PROJECT_DIR/build-support/stage-release.sh v3.1.0-candidate-3 3709463737
```

Then the layout of the current directory will be the same as
https://dist.apache.org/repos/dist/dev/pulsar/pulsar-client-python-3.0.0-candidate-3/.

Fixes #64

It also adds a `upload-pypi.sh` that downloads the wheels from the
Apache's SVN project and upload them to PyPI.

* Add support for Python 3.11

* Removed Python 3.7 windows PR build

* PyYaml

* Use pip3 instead of pip

* Test

* Fixed dependencies wildcard

Fix #84

Release the GIL while calling `producer.sendAsync()` to avoid a deadlock when PyBind is triggering the Python callback. 


  * Main Thread 
     1. Holds the Python GIL
     2. Call `producer.send_async()`
     3. Tries to acquire internal `ClientConnetion` lock

 * Pulsar client internal thread
     1. Holds lock on `ClientConnection`
     2. Receives ack from the broker
     3. Triggers callback
     4. PyBind11 acquires GIL <---- Deadlock

The problem is the different behavior in PyBind from Boost::Python. 

We always need to make sure we release the GIL before making any call to C++ that potentially acquires any mutexes

### Motivation

Adopt the same versioning rule with the Node.js client, see
apache/pulsar-client-node#287.

Add an extra step to commit the version update directly before pushing
the tag. For example, fda5086

(cherry picked from commit cbd31e7)

Addresses Issue #31 - we should access a string representation of the given object rather than assuming that the object itself can be concatenated with a string.

(cherry picked from commit e3eed2d)

(cherry picked from commit ed6983b)

#95)

(cherry picked from commit 2aaacad)

(cherry picked from commit 2bab367)

…terrupted (#99)

### Motivation

Currently, when a blocking API is interrupted by a signal, `SystemError`
will be thrown. However, in this case, `PyErr_SetInterrupt` will be
called and next time a blocking API is called, `std::system_error` will
be somehow thrown.

The failure of
https://lists.apache.org/thread/cmzykd9qz9x1d0s35nc5912o3slwpxpv is
caused by this issue. The `SystemError` is not called, then
`client.close()` will be skipped, which leads to the `bad_weak_ptr`
error.

P.S. Currently we have to call `client.close()` on a `Client` instance,
otherwise, the `bad_weak_ptr` will be thrown.

However, even if we caught the `SystemError` like:

```python
    try:
        msg = consumer.receive()
        # ...
    except SystemError:
        break
```

we would still see the following error:

```
terminate called after throwing an instance of 'std::system_error'
  what():  Operation not permitted
Aborted
```

### Modifications

- Wrap `ResultInterrupted` into the `pulsar.Interrupted` exception.
- Refactor the `waitForAsyncValue` and `waitForAsyncResult` functions
  and raise `pulsar.Interrupted` when `PyErr_CheckSignals` detects a
  signal.
- Add `InterruptedTest` to cover this case.
- Remove `future.h` since we now use `std::future` instead of the
  manually implemented `Future`.
- Fix the `examples/consumer.py` to support stopping by Ctrl+C.

(cherry picked from commit ec05f50)