Clarify installation requirements in quickstart.ipynb

[ArijanaBohr]

Hi! I was trying out the quickstart.ipynb notebook and ran into some installation friction with:

pip install "dpsynth[pipeline,examples] @ git+https://github.com/google/dpsynth.git"

In a Linux environment (Python 3.12), this pulled a very large dependency tree, including many Google Cloud packages, and appeared to stall while resolving/installing dependencies, so I stopped it after about an hour.

I also tried DPSynth on Google Colab (Python 3.13), where installation eventually failed while building pydantic-core; this also happened with the minimal installation, so it may be a separate compatibility issue.

I later noticed that detailed_example_theory_and_in_memory_api.ipynb uses the lighter [examples] installation. For the in-memory API, I tried the minimal installation:

pip install "dpsynth @ git+https://github.com/google/dpsynth.git"

This worked successfully in the same Linux environment, and I was able to generate synthetic data without any issues. :)

Since many users may only need the in-memory API to get started, perhaps the quickstart could begin with the minimal installation/example?

[hanzalaareeb]

Thanks for reporting this. I reproduced the installation behavior on Linux x86_64 with Python 3.12 using DPSynth main.
The minimal installation completed successfully. In contrast, the combined [pipeline,examples] installation entered extensive pip dependency backtracking and did not reach the installation stage during the test.
The current dependency layout helps explain the difference: the examples extra includes dpsynth[text], so [pipeline,examples] combines the Apache Beam/TensorFlow dependencies with the text/Gemma/Google dependency tree, although those packages are unnecessary for the initial in-memory example.

I also tested the minimal installation in a Colab Python 3.12 runtime. Pip selected a compatible prebuilt pydantic-core wheel and completed installation, so I did not reproduce the separately mentioned Python 3.13 pydantic-core build failure.

During testing I encountered the separate duplicate mbi.Dataset registration regression, but that has now been resolved through #202 and its associated PR.
I agree that the quickstart should install dependencies progressively:

1. Begin the in-memory section with:
   pip install "dpsynth @ git+https://github.com/google/dpsynth.git"
2. install dpsynth[pipeline] only before the optional Beam section.
3. Avoid installing the examples/text dependency tree when it is not used by the quickstart.