[CDF-24919] 👷 Fix transformations.preview (#1615)

# Description
The Python-SDK implementation of the `/transformations/run/query` has a
few issues:
1. It is retried. By default up to ten times.
2. It has a client timeout typically set to 30 seconds. That means that
even though you pass in a `timeout=240` in the request body, the client
will timeout before the server timeout.

In the `cdf profile asset-centric` we (mis)use this endpoint to
calculate how many unique metadata keys each asset-centric resource have
(unfortunately `assets/aggregate` cannot do this as it 'standardizes the
metadata keys and thus lose the actual count). In addition, we use it
for doing relationships and label count. These calculation are longer
running calculations for larger projects. What happens for a large
project, as shown below, is that the PySDK tries starts the
calculations, waits 30 seconds (client-side timeout), fails it and does
a retry. This ends up with 10 attempts each 30 seconds long to do this
computations.

This PR overwrites the `transformations.preview(...)` method (the one
that implements `/transformations/run/query`), and forces it to not
retry and sets the client-side timeout to server-side + 60 seconds to
ensure it is the server side that will be the limiting factor. This
enables the profiling of larger projects such as shown below.



![image](https://github.com/user-attachments/assets/1865c1e0-edbd-498a-ae92-102915441085)


## Changelog

- [x] Patch
- [ ] Minor
- [ ] Skip

## cdf

### Fixed

- [alpha] The `cdf profile asset-centric` is much less likely to
time-out when calculating metadata key usage.

## templates

No changes.
