feat: add dependency-free permission-expression evaluator

[angelraph]

Description

Implements a dependency-free, pure evaluator for compound boolean permission expressions, as an isolated primitive independent of the broader GuildPass policy engine.

Adds @guildpass/permission-expression with:

• A strongly typed discriminated-union AST: permission (leaf), all, any, not.
• parsePermissionExpression(input, limits?) — validates untrusted runtime input (not just compile-time-typed input) against the grammar: rejects unknown node types, non-object nodes, empty/non-string permission values, and non-array children. Enforces configurable maxDepth and maxNodes limits, checked before recursing into children, so oversized fan-out or cyclic/self-referential input is rejected with bounded work instead of exhausting memory or the call stack.
• evaluatePermissionExpression(expression, grantedPermissions, { limits?, explain? }) — pure, deterministic, side-effect free. Accepts a Set, array, or any iterable of granted permission strings. No eval, no Function constructor, no dynamic code execution.
• explain: true returns the specific leaves responsible for a denial, including through not (the granted-but-forbidden leaf that triggered the negation).
• Empty all is vacuously true; empty any is vacuously false — both explicitly defined and unit tested.
• Zero runtime dependencies; no knowledge of memberships, Prisma models, Fastify requests, or Redis.

Linked Issue

Closes #348

Type of Change

• 🐛 Bug fix (API or policy engine)
• ✨ New feature / endpoint
• 📝 Documentation / OpenAPI spec update
• 🔧 Chore / refactor / dependency update
• 🧪 Tests only
• ⛓️ Smart contract change (requires extra review)

Changes Made

• packages/permission-expression/src/index.ts — AST types, parsePermissionExpression, evaluatePermissionExpression, PermissionExpressionError.
• packages/permission-expression/src/index.test.ts — 36 unit tests: leaf checks, all/any/not semantics, empty-collection semantics, nested combinations, malformed-input rejection, depth/node-count limits (including a 200k-wide fan-out and a self-referential cyclic node), and explain mode.
• packages/permission-expression/package.json, tsconfig.json — package scaffolding matching existing packages (quorum-engine, rate-limit).
• pnpm-lock.yaml — workspace registration for the new package.

Test Evidence

pnpm --filter @guildpass/permission-expression test

▶ evaluatePermissionExpression - leaf permissions (2/2)
▶ evaluatePermissionExpression - all (2/2)
▶ evaluatePermissionExpression - any (2/2)
▶ evaluatePermissionExpression - not (2/2)
▶ evaluatePermissionExpression - nesting (3/3)
▶ parsePermissionExpression - malformed input rejection (15/15)
▶ parsePermissionExpression - limits (5/5)
▶ evaluatePermissionExpression - explain mode (5/5)

ℹ tests 36
ℹ pass 36
ℹ fail 0

Full workspace (pnpm typecheck, pnpm build, pnpm test) also verified green across all existing packages and apps/api — no regressions.

General Checklist

• I have read CONTRIBUTING.md
• This PR is linked to an open issue
• pnpm typecheck passes
• pnpm build passes
• pnpm test passes — all tests green
• Prisma schema changes include a migration file (N/A — no schema changes)
• New API endpoints are documented in OpenAPI (N/A — no API endpoints)
• No secrets, keys, or wallet addresses introduced
• .env.example updated (N/A — no new env vars)
• Documentation updated if new behaviour was introduced (doc comments in source)

Additional Notes

Deliberately scoped to stay independent of the wider policy engine: no membership lookups, database queries, or access API. Callers supply a set of granted permissions and an expression and receive a result.

[angelraph]

@Lakes41 PR is ready for review — closes #348. All acceptance criteria are met: typed discriminated-union AST (permission/all/any/not), runtime validation with configurable maxDepth/maxNodes limits, deterministic side-effect-free evaluation with explain mode, no eval/Function usage, zero dependencies. pnpm typecheck, pnpm build, and pnpm test all pass workspace-wide (36 new tests, no regressions).

[angelraph]

@Lakes41 Following up — PR #362 (closes #348) has been open with green CI and no review activity yet. Happy to address any feedback whenever you get a chance to take a look.

[angelraph]

@Lakes41 Checking in again — PR #362 (closes #348) is still open with green CI and no review yet. Let me know if anything needs changes.