SRE-1215: Upgrade @graphql-codegen/cli to 7 and its plugins to their next majors - #10088

Requested via Slack thread

🌟 Purpose

Prototype of the GraphQL Code Generator major upgrade, so we can see what it costs. The CLI 6 / plugin 5 line depends on @graphql-tools/utils ^11; the new majors depend on ^12, so this moves @graphql-tools/utils from 11.0.0 to 12.0.3 everywhere in the lockfile.

🔍 Changes

• Versions. @graphql-codegen/cli ^6.2.1 → ^7.4.6, typescript 5.0.9 → 6.1.2, typescript-operations 5.0.9 → 6.1.10, typescript-resolvers 5.1.7 → 6.1.1, introspection 5.0.1 → 6.1.0, fragment-matcher 6.0.0 → 7.1.0. Transitively, the @graphql-tools/* loaders, delegate, wrap, executor (1.x → 2.0.3) and graphql-config 5.1.8. Each package resolves to a single version. @graphql-codegen/add 7.1.1 is added as a dev dependency where it is used.
• Generated file split. typescript-operations 6 now generates the Input and Enum types its operations use, so running it in the same file as typescript declared those types twice (10 to 18 duplicates per file). This follows upstream's multi-file setup: typescript (plus typescript-resolvers in hash-api) writes schema-types.gen.ts, and typescript-operations writes api-types.gen.ts with importSchemaTypesFrom. An add line re-exports the schema types from api-types.gen.ts, so no import site changes. Upstream's single-file setup drops the typescript plugin, but hash-api imports about 100 resolver and args types and the frontend about 20 object types (such as FlowRun) from it, so we still need it.
• base-codegen-config.ts is now typed for both plugins, because skipTypename moved from the shared visitor config to the typescript plugin's config.
• Breaking changes that reached our code. Result types no longer get an optional __typename by default. Apollo Client's InMemoryCache adds __typename to every selection set except the root operation type, so the frontend and hash-isomorphic-utils operation types set nonOptionalTypename and skipTypeNameForRoot to match. hash-isomorphic-utils sets skipTypename: false on its schema types, so they include __typename too. The fetch-based clients (plugin-browser, hash-playwright) keep the default, since nothing adds __typename there. One cast in flow-run-table.tsx no longer typechecked and was dropped. Other changes that applied without type errors: nullable result fields are now T | null rather than optional (274 → 27 optional fields in the frontend's operation types), unmapped custom scalars default to unknown instead of any, and input ID is string | number.
• Removed the unused @graphql-codegen/typescript-graphql-request from hash-backend-integration. Its latest stable release (7.1.0) still depends on the 6.x helpers that pin @graphql-tools/utils ^11, and only an 8.0.0 alpha moves off them.
• Generated code. The *.gen.* files are gitignored, so they are not in the diff. After formatting before and after the same way, the regenerated output changes by roughly +880 / −870 lines across six workspaces (hash-frontend +352/−467, hash-api +264/−154, plugin-browser +96/−99, hash-playwright +92/−99, hash-isomorphic-utils +69/−42, hash-backend-integration +5/−10). The source diff is 93 added and 32 removed lines, plus the lockfile.

Checked locally:

• codegen in all six codegen workspaces.
• lint:tsc through turbo for hash-isomorphic-utils, hash-backend-utils, hash-api, hash-frontend, plugin-browser, hash-playwright and hash-backend-integration. All clean.
• ESLint and oxfmt on the changed files.
• test:unit for hash-isomorphic-utils (360 tests), hash-api (311) and hash-frontend (1450). All pass.
• yarn install --immutable, yarn dedupe --strategy highest --check and yarn constraints, all with Yarn 4.18.1.

⚠️ Known issues

• Not run locally: integration tests, Playwright, and a manual check of the frontend in a browser. The type changes are compile-time only, but the lost ? on result types could hide a mismatch with what Apollo's cache returns at runtime.
• The split is the smallest change for consumers. Moving the frontend and other clients to the single-file typescript-operations setup that upstream recommends would mean replacing their uses of schema object types, and it is out of scope here.