CDK: Improve schema detection

[flash1293]

What

Closes #25669

Improves the built-in schema inferrer to produce schemas that work better with the Airbyte platform by default:

• Always assume a number is "number", not "integer", because the sample size is usually small
• Ignore properties of type "null", as this is most likely not correct anyway and can cause errors later on when the schema gets refined - instead this will just be unexpected fields that are filtered out by the platform
• Do no emit types like this:

{
  "a": {
    "anyOf": [
      { "type: "null" },
      { "type: "object", "properties": { ... } },
    ]
  }
}

instead, do this:

{
  "a": {
    "type": ["null", "object"],
    "properties": { ... }
  }
}

as the platform is ignoring anyOf but can extract the nested object in the second case.

How

The number/integer case can be handled by adding another extra strategy.

However, the null handling is baked in too deeply into genson to be able to get the desired behavior by extra strategies (e.g. the anyOf can't be changed by provided strategies and the library always expects an output for a schema node, so an "ignore null" strategy can't be implemented). To solve this, a post-processing function is introduced that traverses the built schema and changes the output as desired.

🚨 User Impact 🚨

Schema detection in the connector builder will change for the better (existing connectors won't be affected)