docs: define memory benchmark scenarios

[anusbutt]

Outcome

Adds benchmarks/graphkeeper-memory-bench-v0.1.md with the first four scenarios for evaluating durable project memory in coding agents. The change is limited to scenario definitions; scoring, datasets, automation, implementation, tests, thresholds, and additional scenarios remain out of scope.

Evidence

• git diff --cached --check passed before commit.
• Tests were not run because this is a documentation-only addition with no runtime behavior changes.

Checklist

• Tests cover valid, invalid, and relevant security or performance paths. (Not applicable: no implementation or tests were requested.)
• Documentation and examples reflect every changed public behavior.
• Schema compatibility is unchanged, additive and documented, or explicitly marked breaking.
• Constitution principles and append-only/evidence guarantees remain satisfied.
• npm run build passes. (Not run: documentation-only change.)
• npm run typecheck passes. (Not run: documentation-only change.)
• npm test passes with no hidden skips. (Not run: documentation-only change.)
• npm run package:smoke passes and the tarball contains intended assets only. (Not run: documentation-only change.)
• sh -n scripts/validate.sh and sh -n templates/pre-commit pass when shell files change. (Not applicable: no shell files changed.)
• No secrets, credentials, telemetry, generated archives, or unrelated edits are included.

Compatibility and recovery

No runtime platforms, public commands, schemas, diagnostics, or dependencies change. Rollback is a straightforward revert of the documentation commit; no manual recovery is required.