Refactor the data model of pytest metadata

[ono-max]

Before

"data": {
  "markers": [
    { "name": "name", "value": "foo" },
    { "name": "args", "value": "()" },
    { "name": "kwargs", "value": "{}" }
  ],
  "lineNumber": 3
}

"data": {
  "markers": [
    { "name": "name", "value": "parametrize" },
    { "name": "args", "value": "('y', [2, 3])" },
    { "name": "kwargs", "value": "{}" },
    { "name": "name", "value": "parametrize" },
    { "name": "args", "value": "('x', [0, 1])" },
    { "name": "kwargs", "value": "{}" }
  ],
  "lineNumber": 12
}

After

"data": {
  "markers": [{ "name": "foo", "value": "" }],
  "lineNumber": 3
}

"data": {
  "markers": [
    { "name": "parametrize", "value": "('x', [0, 1])" },
    { "name": "parametrize", "value": "('y', [2, 3])" }
  ],
  "lineNumber": 12
}

[ono-max]

c.c. @NicuPascu

[kohsuke]

Thanks for taking care of this so quickly!

• In Python's function calling convention, arguments can be partially positional (what Python calls args) and partially keyword based (what Python calls kwargs), for example f(1,2,a=3) so assuming this metadata follows the same scheme, this code could see both of them at the same time. If that happens, this simply drops the positional arguments, which would be hard to notice. If you choose to handle this, I suppose the easiest is to just concatenate both args and kwargs. Python invocation syntax (1,2,a=3) would be probably nicer in terms of visualization, but it's bit more effort processing. Maybe Claude Code could write that for us?
• I believe the test dependency mechanism that some customers rely on are built on this metadata. If so, I'd imagine we need to update the server side first to be able to deal with both forms, before we roll out this change.

[NicuPascu]

LGTM 🐝

Can we go through the DB and update the format of the all the old markers as well or should we support both data structures?