fix python profile html generator to support updated JSON format.

[ChristianMurphy]

when a profile is generated using

-- enable profiling in json format
PRAGMA enable_profiling='json';
-- write the profiling output to a specific file on disk
PRAGMA profile_output='/path/to/file.json';

and an HTML document is generated using

import duckdb_query_graph
duckdb_query_graph.generate('/path/to/file.json', '/path/to/out.html')

as documented at https://duckdb.org/docs/dev/profiling

Two issues occur.

1. graph_data is now the root node of the graph, it is no longer nested
   under graph_data.tree
2. extra_info is now option, it is present on some, but all nodes

this addresses both issues, but starting the parsing from the root node
directly, and adding an empty fallback for extra_info so split() does
not crash.

[Mytherin]

Thanks for the PR!

It was not entirely clear to me, the problem here is that the graph generation fails without these changes, correct?

Could you perhaps also add a test case to the CI that tests this workflow so that this does not break again? For example this could be added to the Python client tests, or as a shell script to e.g. the LinuxRelease CI run.

[ChristianMurphy]

It was not entirely clear to me, the problem here is that the graph generation fails without these changes, correct?

Correct, more specifically.
The HTML and JavaScript generated by Python is valid (parse-able).
But when it is run in a browser, the graph generation function on the JavaScript side fails with various undefined exceptions, resulting in a blank page in browser.

For example:

For example this could be added to the Python client tests

I'm not sure a Python client test would be able to catch this.
Unless, a browser was installed as part of CI, and the Python tests use a driver to instruct the browser to open the generated HTML+JS document.

Which is possible with libraries like https://playwright.dev/python/docs/intro and https://selenium-python.readthedocs.io/, but would add the overhead of needing to download browser(s) in the CI task and spin them up for this one test.

[Mytherin]

Fair enough, I thought the problem would manifest already in the actual Python generation, but if the problem is in the generated javascript this is indeed much more difficult to test. I will just merge the PR as-is then.

Cheers!