fix(graphql): add deleted_classes:[UserDetails] migration to unblock deploys

[mattzcarey]

Problem

The staging and production deploys of #384 (which removed the UserDetails Durable Object) failed, leaving most servers undeployed:

• graphql → code: 10064 — "New version of script does not export class 'UserDetails' which is depended on by existing Durable Objects … use a delete-class migration."
• workers-observability → code: 10061 — "Cannot apply --delete-class migration to class 'UserDetails' without also removing the binding that references it."

Because turbo deploy aborts on first failure, graphql's failure halted the whole release.

Root cause

#384 removed the UserDetails DO and its bindings from every server, and correctly added a deleted_classes: ["UserDetails"] migration to the two servers that own the class (workers-observability, workers-builds). It did not add one to graphql, whose already-deployed worker still depends on UserDetails — so wrangler deploy refuses the new version (10064). graphql is also the last consumer still holding a cross-script binding to observability's UserDetails, which is what blocked observability's own delete-class migration (10061).

Fix

Add the delete-class migration to graphql:

"migrations": [
  { "new_sqlite_classes": ["GraphQLMCP"], "tag": "v1" },
  { "deleted_classes": ["UserDetails"], "tag": "v2" }
]

graphql owns no actual UserDetails namespace (only observability does), so this just clears graphql's dependency on the class and releases the binding that was blocking observability.

Validation

Deployed the full set to staging in order (consumers first, owners last):

• graphql-staging and workers-observability-staging now deploy cleanly (10064 / 10061 gone).
• All staging servers respond 200 on the new build (accounts_list tool gone).
• The staging UserDetails Durable Object namespace is deleted; only the production one remains (pending the prod redeploy).

Follow-up / deploy note

Production graphql + workers-observability are still on the pre-#384 build (their failed deploys left the old version running — no downtime, UserDetails data intact). After this merges, redeploy them to production in order (graphql → observability) so observability's delete-class migration succeeds.