This is not the optimal way to address #1047 but is the most backwards
compatible.

includes also changes to transform method arguments
see #1614

* implicit pruning from healSchema and mapSchema can sometimes be dangerous: e.g.  empty interfaces are valid for later extension.
* the pruneSchema utility functions (and PruneSchema transform) can be explicitly used to prune as desired

See: #1817

* removes support for createMergedResolver and ExtendSchema transform in favor of visitResult, simplifying the refactoring.

move typesContainSelectionSet to delegate package
remove typeContainsSelectionSet

Previously, errors with invalid or missing paths were lost.
See:
#1641
https://github.com/apollographql/apollo-server/issues/4226

They are now saved!

All correctly pathed errors are now inlined into the returned data by the CheckResultAndHandleErrors transform.

The data is now annotated only with the remaining "unpathed" errors. These are then returned when possible if a null is encountered including the missing or potentially invalid path.

Changes to error handling obviate some existing functions including getErrorsByPathSegment, getErrors in favor of getUnpathedErrors.

Utility functions including slicedError and unreleased functions extendedError and unextendedError are no longer necessary.

we could consider later adding support for $$asyncIterable similar to upstream graphql-js, but this would have to be across the entire codebase, which currently supports only Symbol.asyncIterator.

use delegationContext within delegation transforms
remove fragment hints and transforms in favor of selection set hints and transforms
prune fragment functions from utils package

* updated filter schema.

* update existing usage.

#2062

= remove schemaLikeObjects in favor of better typed sources

= make sure to retrieve the transformedSubschema stored with the transformedSubschemaConfig

requires passing additional parameters to each transformSchema call beyond the subschemaConfig: the transforms, and the transformedSchema.

Because transformSchema may need to create additional proxying resolvers, it needs all the parameters that would be passed to delegateToSchema.

It is NOT a catch-22 that transformSchema requires the transformed schema as a parameter, because it does NOT require an executable transformedSchema as a parameter, only the non-executable version for type comparisions performed ultimately by delegateToSchema within the proxying process.

This does not make sense in the context of planned functionality to wrap subscription root fields in potentially multiple layers, as the wrapping resolvers will have to be different for the root field and the other layers.

Changing a root field resolver can be easily accomplished in a separate transform via TransformRootFIelds or TransformObjectFIelds.

Remove `Operation` type in favor of `OperationTypeNode` from upstream graphql-js

resolvers that return 'subresolvers' or fields that are set to functions for later use as resolvers were behaving not according to upstream graphql-js convention, these functions take 3 arguments, not 4, with parent available as this.

See #1807
See https://github.com/graphql/graphql-js/blob/7e79bbe5f2b0e971b5e5f6fe3e7b19c43dea9f35/src/execution/execute.js#L1210-L1212

includes breaking changes to onTypeConflict, fieldConfigMerger and inputFieldConfigMerger

… array

see #2103

to allow batching when using the default executor

BREAKING CHANGE: batching is now performed (when enabled) for all subschema config objects with identical executors, removing the need for the endpoint abstraction