schema: declare non-null elements on eight list positions

[dkijania]

Closes #236.

Eight list positions go from [T]! to [T!]! (eventData/actionData: element only, the field stays nullable). No resolver produces a null element; see #236 for the per-position evidence.

• graphql-inspector: 8 changes, no breaking change.
• tests/unit/schema-nullability.test.ts: every output list must declare non-null elements, except ZkappFieldArray.fields. Fails on main with exactly these eight, passes here.
• Unit 216/216, integration 91/91 (Node 24, PostgreSQL 17); no Cannot return null in any response.

Lets the three SDKs keep non-null element types in their 2.0 instead of widening them.

🤖 Generated with Claude Code

https://claude.ai/code/session_01StPY7C6STPhukrgZvVhqZX

[SanabriaRusso]

Approving. All eight positions are output-only, so nothing changes on the wire for o1js, mina-explorer or the client SDKs, and no input type or argument is touched.

What I verified

• graphql-inspector@7.0.0 (the version CI uses), run against main: 8 changes, all safe, "No breaking changes detected".
• No position can return a null element. I traced each one to the code that produces it:
  • Query.events, Query.actions and Query.blocks are object literals built in the services, with ?? [] defaults.
  • EventData.data / ActionData.data come from zkapp_field.field, which is text NOT NULL, through an INNER JOIN. In removeRedundantEmittedFields, .flat() drops the sparse holes.
  • zkappAccountUpdateIds comes from zkapp_account_updates_ids, which is integer[] NOT NULL.
• The wider null spread under [Block!]! can't be reached today. Every required Block sub-field is safe:
  • creator comes from an INNER JOIN.
  • state_hash and parent_hash are NOT NULL.
  • coinbase and amount are COALESCEd.
  • A bad timestamp already failed the whole query before this PR.
• Codegen. npm run codegen reproduces src/resolvers-types.ts exactly.
• Tests.
  • Unit tests: 216/216.
  • Integration tests: 91/91 on Postgres 17.
  • Live GraphQL queries for events, actions and blocks(limit: 500) against the fixture DB returned data with no "Cannot return null" errors.
• CI. All checks pass.

Optional

• Null spread note. Issue #236 asked for one sentence in the description about the wider null spread for blocks: a future computed field that throws would null the whole result. Worth adding so the note reaches the squash commit.
• SCHEMA_VERSION. It stays 1.1 here. #245 moves it to 2.0 on top of this PR, so no bump is needed in this PR.