To support standalone stage initializers (such as subcollection()), that can be constructed without an existing pipeline, we need to decouple stage initialization and user data validation. Standalone functions do not have access to Firestore or databaseId context when called, so validation must be deferred until the pipeline is executed with execute(pipeline).

Implementation Details:

1. Removed user data parsing in stage methods (select, where, addFields, ...).
2. Added a _validateUserData() traversal to Pipeline class and all stages.
3. Modified execute() to trigger pipeline._validateUserData() before executing the pipeline.

Breaking Change:
Errors regarding invalid user data in stages will now be thrown when execute(pipeline) is called, rather than during the initialization of the stage (pipeline.collection().addFields(/* invalid data */)).
This aligns with the behaviour on both iOS, Android, and Java SDKs.

// Previously: error threw here
const ppl = firestore.pipeline().collection('books').select(field, undefined); // undefined is an unsupported field value

// Now: error throws here
await execute(ppl);

Ported from: firebase/firebase-js-sdk#9715

I also updated the ESLint config rules to allow args to be unused if their identifier starts with an underscore (_name). Previously the rule was that they can only be underscores (_), which causes issues if there's more than one unused argument. I ran into this issue when implementing the _validateUserData(name: string, ignoreUndefinedProperties: boolean) methods.