test: cover unary cursor pagination projections

[shivros]

Summary

Adds a focused Hydra codegen regression fixture for one explicit unary ContextPage operation with query limit and opaque cursor inputs.

• Verifies CLI, HTTP, and MCP projection from the same definition.
• Verifies MCP query-location metadata and HTTP query forwarding.
• Verifies byte-identical CLI, HTTP, and MCP artifacts across two generate_all calls.
• Documents that cursor retention/replay and next_cursor semantics remain consumer-owned.

Validation

• cargo test -p hydra-codegen
• cargo test --workspace --all-targets
• cargo clippy --workspace --all-targets -- -D warnings
• cargo fmt --all -- --check

Closes COD-454.

[shivros]

🤖 Automated Review Panel

Reviewer A — GPT-5.6-terra

APPROVE. The explicit unary get_context_page fixture covers the declared CLI/HTTP/MCP allowlist, query extraction and forwarding, MCP schema/location metadata, and byte-identical repeated generation. It introduces neither SSE nor consumer pagination semantics.

Reviewer B — independent GPT-5.6-terra review

APPROVE. Independently verified the current worktree: the fixture uses one declared operation and asserts all required surfaces plus deterministic artifacts. No correctness or goal-alignment defect found.

The cron environment returned the inherited GPT-5.6-terra model for both review invocations; Gemini/OpenRouter was not available to this run. This is recorded rather than misattributed.

Both reviewers confirmed the full Rust gates were clean.

[shivros]

Auto-Merge Gate

Confidence: 0.94

The single test-only change directly fulfills COD-454: one explicit unary ContextPage definition is projected through CLI, HTTP, and MCP with query-location preservation and deterministic repeated generation. It deliberately keeps cursor replay and next_cursor domain semantics outside Hydra.

Evidence:

• GitHub CI Test and Creed context drift both succeeded.
• Local verification: cargo test -p hydra-codegen passed (38 integration tests, including the new cursor-pagination projection test).
• Existing automated review panel approved both reviews; no formal or inline blocking feedback exists.
• Scope is one 95-line test file; no runtime, deployment, credential, or destructive behavior changes.

Squash-merge is appropriate; COD-454 acceptance criteria are satisfied by this PR.