Add regression test for RelayResponseNormalizer @skip + multi-fragment warning (#5119)

[markselby9]

Summary

Adds a regression test in RelayResponseNormalizer-test.js for the bug reported in #5119: when two fragments in the same query select the same field but only one of them gates the selection with @skip(if: $skip), the normalizer emits the warning Payload did not contain a value for field ... without identifying the originating query.

This is test-only — no production code changes, no improvement to the warning message format. The test asserts current (buggy) behavior; whoever fixes the underlying bug or improves the warning to include source-query attribution should flip the assertion.

The reporter called out __typename as the painful case ("especially difficult to debug"). The test uses firstName instead — a __typename variant doesn't trigger the warning here because the parent selector ... on User requires node.__typename to be present in the payload, which short-circuits the warning path before it reaches the User fragments. The in-test comment explains why firstName is path-equivalent: both traverse _normalizeField's scalar warning path with no abstract-type discrimination involved when the parent (User) is a concrete type.

The test reuses the existing convention from the file (expectToWarn, RelayRecordSource, createNormalizationSelector, disallowWarnings() in scope at the top). Both fragment spreads carry @dangerously_unaliased_fixme because the compiler now requires aliases on duplicate selections; this matches the convention used elsewhere in the same file (e.g. the _pvFragment test).

Refs #5119.

Notes for reviewer

• Generated artifacts: the test introduces a new query and two new fragments, requiring 3 new __generated__/*.graphql.js files. They were produced with relay-compiler@0.0.0-main-38ad9e13 (the published @20.1.1 rejects fields the current OSS test config uses). The structure follows the standard codegen format; if the maintainer's internal compiler regenerates with trivial differences (SignedSource hash, formatting), please regenerate as part of the import — happy to follow up if it causes any review friction.
• Test name is intentionally bug-leading ("warns without identifying source query…") rather than trigger-leading, so the regression intent is clear at a glance.
• Assertion text is generic. The expectToWarn(...) matcher is the same generic missing-field warning the existing L2640 test exercises with a trivial single-fragment query. The regression value of this test is in the query/fragment shape (sibling fragments, one @skip-gated), not in the assertion text. When the underlying bug is fixed (either better source attribution OR no-warn-on-this-pattern), the assertion needs to be updated — please cross-reference this comment if doing so.
• Scope discipline: single new it() in the existing describe('RelayResponseNormalizer', ...) block, immediately after the existing 'warns in __DEV__ if payload data is missing an expected field' test (sibling concept).

Commits

• Commit 1: regression test + 3 generated artifacts
• Commit 2: dropped unused const SkipFragment/const NoSkipFragment assignments to satisfy eslint --max-warnings 0 (the fragments are picked up by the compiler from the tagged template literal regardless of whether the value is assigned)

Test plan

• yarn jest packages/relay-runtime/store/__tests__/RelayResponseNormalizer-test.js -t '5119' — 1 passed
• full RelayResponseNormalizer-test.js suite still passes (61 total, was 60)
• yarn run lint packages/relay-runtime/store/__tests__/RelayResponseNormalizer-test.js — clean (0 warnings)