Tying to fix issues on AI/BI and Genie skills

[MigQ2]

I spent some time trying to get an agent to make a Genie space and a dashboard and it struggled with the following:

AI/BI:

1. For multiple iterations, all charts showed an error due to a missing GROUP_BY, this seems to be related to management of disaggregated
2. The agent didn't know clearly how to apply a filter to multiple datasets

Genie:

• I noticed the agent was only able to edit the basic settings (i.e. name and description) of the Genie using the skill and MCP tools. However the API exposes a serialized_space parameter that can be used so that agents can also edit other Genie properties like joins or benchmarks

IMPORTANT NOTE

This PR was written by the agent (which was using Claude Sonnet 4.6) based on what it learned during these development iterations until I managed to get a working high quality dashboard and Genie.

I am not an expert in the underlying serialization of dashboards and Genie spaces. It would be great if this could be reviewed by someone who knows well the internals of both and can refine it to make sure everything is following the right design pattern

Also would be nice if the MCP server would already include serialized_space so that agents would't need to use the API

[calreynolds]

@MigQ2 #233 is about to be merged! Would you want to tack on changes on top of that? \ merge those changes into this PR then strip away the serialized spaces stuff 👍

(Then would love to have your optimizations that I see here)

[calreynolds]

per comment (shifting to review 👍 )

[MigQ2]

@calreynolds TBH I don't feel confident editing or merging this.

The content of this PR is what an agent figured out by itself trying to make dashboards and genie work.

I believe the right way should be someone from Databricks who deeply understands the underlying serialization formats for Genie and AI/BI checks this out and curates it because those formats are not publicly documented in detail.