Use execute_query for introspection

[codedmart]

What

The configuration introspection was using select_first_row which was not properly streaming all results and had a limits which caused introspections to fail on larger sets of tables.

How

This removes select_first_row and uses execute_query to fetch all introspection data.

[daniel-chambers]

I don't feel like I understand what's going on here with this Vec and execute_query. From my understanding there is only one row returned by the introspection query... but the way this is implemented, each row's string is just blindly concatenated into the same array of chars here (stored_procs_row is Vec<u8>). That seems... dubious.

My quick reading of the tiberius API we're using is that we should be taking a QueryStream and calling into_first_result, which gives a Vec of Row. Then we should extract our JSON from the one column in the one row we find there and error if we get more than one row because why are our JSON queries returning multiple rows...

The fact that our logic reads multiple rows and munges all data from the first column in each row into a single string is scary. I can't find anything my quick check of the SQL Server docs or AI queries that indicate that a JSON string would be split across multiple rows which is what the execute_query code seems to be saying will happen.

Edit: Is it because of this? https://dba.stackexchange.com/a/279614 If so, let's document that inside execute_query and rename our variables to not refer to "rows" since there's really only one and the butchered "rows" sent over the wire is just some crazy XML streaming stuff :/

Also maybe let's refactor execute_query to just return the string rather than having a mutable buffer passed in... I can't see the reason for wanting that sort of mutable API for this rather than a straight functional one. We can do that separately, of course.

[codedmart]

I'll create a ticket to do this in a follow up. Good call out.

[codedmart]

Created ticket here: https://linear.app/hasura/issue/ENT-152/document-and-refactor-execute-query-in-ndc-sqlserver

[danieljharvey]

This is a big improvement, let's merge this and do the other identified improvements in a follow up. You'll need to run just format to get rid of the cargo fmt errors.