Pyvespa

* Add retry strategy to `delete_data`, `get_data` and `update_data` (https://github.com/vespa-engine/pyvespa/pull/222)
* Deployment parameter `disk_folder` defaults to the current working directory for both Docker and Cloud deployment (https://github.com/vespa-engine/pyvespa/pull/225)
* Vespa connection now accepts cert and key as separate arguments. Using both certificate and key values in the cert file continue to work as before. (https://github.com/vespa-engine/pyvespa/pull/226)

* Infer schema name whenever possible for batch document operations (https://github.com/vespa-engine/pyvespa/pull/210)
* When we have application package information containing just one schema, it is possible to simplify the batch document operations by not specifying the schema name. For those cases, we can for example use app.feed_batch(docs) instead of app.feed_batch(docs, schema = "sentence") when we know there is only one schema named "sentence".
* Introduce TextSearch use case (https://github.com/vespa-engine/pyvespa/pull/212)
* Create a convenience function to feed df to a Vespa app (https://github.com/vespa-engine/pyvespa/pull/213)
* Allow a data frame to be used when collecting training data (https://github.com/vespa-engine/pyvespa/pull/215)
* Allow default query model to be specified and define it for TextSearch (https://github.com/vespa-engine/pyvespa/pull/217)

Removes `vespa.package` module dependency on the `vespa.ml` module. The main goal is to avoid installing `vespa.ml` specific dependencies through `pip install pyvespa[ml]` unless the user explicitly wants to use `vespa.ml` classes.

* Use `tasks` instead of `models` as argument name for `ModelServer` to align with the fact this argument takes instances of type `TextTask`.
* Use `model_id` instead of `model_name` to identify a model since `model_id` is what we use when defining a Task.

Introduce stateless model evaluation with SequenceClassification task.
* Adapt `ApplicationPackage` to allow for simpler configurations, e.g. no schema.
* Implement `ModelServer`, which is a simplified `ApplicationPackage` focused on stateless model evaluation.
* `ApplicationPackage` is now included on the `Vespa` instance when deploying apps with pyvespa.
* This helps to process input/output involved when using stateless model evaluations through `app.predict`.
* `Vespa` now has `get_model_endpoint` and `predict` to use for stateless model evaluation mode.
* Both `VespaDocker` and `VespaCloud` now work with `ModelServer` deployment to deploy stateless model evaluation apps.
* Created `SequenceClassification` task that inherits from the base class `TextTask`.

VespaDocker` and `VespaCloud` are now only available via the `vespa.deployment` module.