refactor query graph and enable profiling for a connection

[guillesd]

The idea of this PR is to make this tool a bit more useful (and visually appealing). I think generally it is good to have some better profiling tools to explore queries.

EDIT: Initially it was only about the query graph visuals, but I decided to expose the profiling functions enabled in the C++ API (similar to what the Go client does with the C API).

To use the tool:

1. Build this branch (see https://duckdb.org/docs/stable/dev/building/python)
2. Run the following in DuckDB

PRAGMA enable_profiling = 'json';
PRAGMA profiling_output = './tmp/profile.json';
SELECT ... FROM ...;

3. Run the script (I do it with uv):

uv run -m duckdb.query_graph --profile_input profile.json

If you want to use the profiling within the client:

import duckdb
from duckdb.query_graph import ProfilingInfo

con = duckdb.connect()
con.enable_profiling()
con.execute("select 42").fetchall() #or some other eager operation
profiling_info = ProfilingInfo(con)

# then the options are
profiling_info.to_json()
profiling_info.to_pydict()
profiling_info.to_html()

# optionally
con.disable_profiling()

It is also possible to just provide the profiling info as a file:

# we create profiling info from a file instead of binding it to a connection
profiling_info = ProfilingInfo(from_file = 'profile.json')

# then the stuff
profiling_info.to_html(output_file='profile.html')

Which should yield something like the following screenshot: