Full e2e system tests

## Problem

The SDK has unit tests with mocked dependencies but no end-to-end system tests that exercise the full flow against real contracts. Encoding bugs, approval flows, recipient handling, and provider routing are never validated on-chain.

## What we need

### Full matrix coverage

Test every combination a developer would encounter:
- **Every wallet provider** (Privy, Turnkey, Dynamic, SmartWallet) × **every chain** (Mainnet, OP Mainnet, Base) × **every namespace function** (getQuote, execute, getMarkets, getMarket, getPosition, openPosition, closePosition)
- **Every swap provider** (Uniswap, Velodrome/Aerodrome) across their supported router types and pool types
- **Every lend provider** (Aave, Morpho) including native ETH paths (WETHGateway) and ERC20 paths
- Both namespace paths: ActionsNamespace (read-only, no wallet) and WalletNamespace (full execution)

### Real execution, no mocks

- Use Anvil forks against Mainnet, OP Mainnet, and Base
- Create real ActionsConfig instances with real providers and real ChainManagers
- Use funded test accounts from Anvil
- Verify on-chain state after execution: token balances, allowances, receipts, recipient addresses

### Test structure and shared state

- Use `beforeAll` / `afterAll` hooks to set up expensive resources (Anvil forks, funded accounts, provider instances) once per suite, not per test
- Tests within a suite should cascade context — e.g., a quote test produces a quote that the execute test consumes, an openPosition test creates state that getPosition and closePosition verify
- Build helper functions and abstractions to minimize per-test boilerplate — e.g., `expectBalanceChange(token, address, fn)`, `withAnvilFork(chain, fn)`, `createFundedActions(chain)`
- Reduce actual test code as much as possible — the matrix should be driven by data (arrays of providers, chains, assets) rather than duplicated test bodies
- Shared utilities in a `setup.ts` or `helpers.ts` that all system test files import

### Edge cases worth covering

- Quote from Actions namespace executed through Wallet namespace (recipient mismatch → re-encoding)
- Existing allowances (verify no redundant approval txs)
- Native ETH swaps
- Quote expiration (advance Anvil time)
- Blocklist enforcement on getMarkets

### Expect to find real bugs

This work will likely surface real bugs in encoding, approval flows, or edge cases. The scope of this issue is **setting up the test matrix and CI job, not fixing the bugs**. When a test reveals a real issue:
1. Write the test anyway
2. Skip or comment out the failing test with a clear note
3. Create a separate issue describing the bug
4. Link the issue in a `// TODO(#xxx)` comment next to the skipped test

Do not block this PR on fixing discovered bugs.

### Infrastructure

- Separate vitest project with extended timeouts
- Separate CI job (see #332, #333) with Anvil and RPC env vars
- Randomized test order

## Existing foundation

`VelodromeSwapProvider.network.test.ts` already has Anvil fork tests for quoting. Expand from there.

## Related

- #332 — Add network fork tests to CI
- #333 — Parallelize tests / split CI into per-package jobs

Start by reading and running VelodromeSwapProvider.network.test.ts, the existing Anvil fork foundation. Then review #332 and #333 for CI and test-splitting requirements before mapping the provider, chain, namespace, swap, and lending matrix into shared system-test setup. Done means the fork-based suites, state verification, edge cases, and separate CI job are in place, with discovered bugs tracked separately.

Written by the indexing model from the issue text.