<!-- REMOVE IRRELEVANT COMMENTS BEFORE CREATING A PULL REQUEST -->
## Changes
<!-- Summary of your changes that are easy to understand. Add
screenshots when necessary -->

### Linked issues
<!-- DOC: Link issue with a keyword: close, closes, closed, fix, fixes,
fixed, resolve, resolves, resolved. See
https://docs.github.com/en/issues/tracking-your-work-with-issues/linking-a-pull-request-to-an-issue#linking-a-pull-request-to-an-issue-using-a-keyword
-->

Resolves #..

### Tests
<!-- How is this tested? Please see the checklist below and also
describe any other relevant tests -->

- [ ] manually tested
- [ ] added unit tests
- [ ] added integration tests
- [ ] verified on staging environment (screenshot attached)

## Changes
The `make_serving_endpoint` fixture [stopped working as of
yesterday](https://github.com/databrickslabs/pytester/actions/runs/13405748534).
Apparently, the created models from the `make_model` fixture do not come
with a model version anymore (which used to be `'1'`).

The `make_serving_endpoint` is updated to fallback on a UC model
version. Also, it allows users to provide input parameters in case they
want to use another model.

### Linked issues

Resolves #databrickslabs/ucx#3714
Resolves #databrickslabs/ucx#3715

### Tests

- [x] added unit tests
- [ ] fixed integration test

* Let `make_serving_endpoint` reference a valid model version
([#106](#106)). In this
release, the `make_serving_endpoint` fixture has been updated to enhance
its functionality and flexibility. The fixture now accepts optional
keyword arguments for specifying the endpoint name, model name, and
model version, with the model version defaulting to `1` and the latest
version automatically retrieved for workspace local models. The served
model input is created with the specified model name, model version, and
workload size, and a mock endpoint object is returned if obtaining the
served model fails. The `remove` method has been updated to accept a
`ServingEndpointDetailed` object for deletion, and the fixture now
maintains a logger instance for warning messages during model version
retrieval. Unit tests have been added to ensure the changes work as
expected, although integration tests have not been fixed yet. These
updates address issues arising from changes in the `make_model` fixture
and improve the robustness and flexibility of the
`make_serving_endpoint` fixture.

## Changes

Set upper SDK limit to 1.0 as this is a library and the projects using
pytester should set the right SDK limit

---------

Co-authored-by: Guenia Izquierdo Delgado <guenia.izquierdo@databricks.com>

* Set upper SDK limit to v1
([#109](#109)). In this
change, the upper SDK limit is set to version 1.0 in the pyproject.toml
file of a library project, allowing projects using this library to set
their own appropriate SDK limits. The databricks-sdk dependency is
updated from "<0.42" to "<1.0" as a result of this commit. Additionally,
the `create` function in the `ml.py` file of the pytester library has
been modified. Previously, the `EndpointCoreConfigInput` object was
passed as an argument to the `serving_endpoints.create` method using the
keyword argument `served_models`. Now, `EndpointCoreConfigInput` is
passed as a keyword argument `config`, which contains the
`served_models` parameter, aligning with the new SDK version 1.0 and
ensuring consistent endpoint creation. These changes are intended to
ensure compatibility with a variety of SDK versions and improve the
library's usability for software engineers.

## Changes

[Databricks SDK
0.51](https://github.com/databricks/databricks-sdk-py/releases/tag/v0.51.0)
included some breaking changes, including removal of an enumeration
where a set of well-known string constants must now be used instead:
> Breaking] Removed large, medium and small enum values for
databricks.sdk.service.serving.ServedModelInputWorkloadSize.

The change is not forward-compatible, so in addition to updating the
fixture we need to bump the minimum-required SDK version because older
versions expect the enum.

### Credits

This was earlier discovered and reported by @cornzyblack as #156.

### Linked issues

Resolves #156.
Supersedes #157.

This is also causing CI test failures on downstream projects including:

 - [databrickslabs/lsql](https://github.com/databrickslabs/lsql)
-
[databrickslabs/blueprint](https://github.com/databrickslabs/blueprint)

### Tests

- existing unit tests

* Compatibility fix for Databricks SDK 0.51
([#158](#158)). The
minimum required version of the Databricks SDK has been updated to
0.51.0 to accommodate breaking changes introduced in this version,
including the removal of certain enumeration values. Notably, the
`ServedModelInputWorkloadSize` enumeration has been replaced with
well-known string constants, prompting an update to the `create`
function to use the string constant `Small` instead of the removed
`ServedModelInputWorkloadSize.SMALL` value. This change affects the
creation of serving endpoints, which now specify workload size using a
string value, and resolves issues reported in related projects,
including test failures in downstream projects, by ensuring
compatibility with the updated SDK version.

Internally the unit tests for pytester access pytest fixtures directly.
This has been [deprecated since
4.x](https://docs.pytest.org/en/stable/deprecations.html#calling-fixtures-directly),
but the project currently contains some workarounds (in
`call_fixture()`) to do this anyway.

In pytest 8.4, the internals changed which breaks the way
`call_fixture()` accesses the underlying fixture. This PR:

 - Updates `call_fixture()` minimally so that it also works with 8.4+.
 - Adds a marker that we should fix this properly.
- Notes a promising direction for doing what we want in a supported
manner.

Resolves #165.