fix(eth): read handlers participate in failover; method-rejection counted as transient

[BitHighlander]

Symptom (CowSwap)

```
| ethereumHandler | | getProvider | | Trying RPC [1/25]: https://rpc.flashbots.net/
| ethereumHandler | | getProvider | | ✅ RPC working! Block: 24997148
| handleWalletRequest | Error processing method eth_call: server response 403
responseBody: {"jsonrpc":"2.0","error":{"code":-32601,"message":"rpc method is not whitelisted"}}
```

Repeated on every subsequent eth_call — Flashbots gets picked first every time.

Root cause

Flashbots' RPC is narrow-purpose (privacy/MEV protection): it serves `eth_sendRawTransaction`, `eth_chainId`, and `eth_blockNumber`, and rejects everything else with HTTP 403 + JSON-RPC `-32601` (`rpc method is not whitelisted`). Pioneer's catalog ranks it in the top tier (correct for its actual purpose), so:

1. `getProvider()` picks it first
2. The pre-flight `getBlockNumber()` test passes (it's one of the supported methods)
3. `getProvider()` returns the Flashbots provider
4. The actual handler call (`eth_call` / `eth_getBalance` / `eth_getCode` / `eth_estimateGas` / etc.) hits 403 → throws → handler dies
5. `failedRpcs` cooldown is never set because the cooldown is only updated when the pre-flight test fails, not when the actual call fails
6. Next request: same flow, same Flashbots, same 403

Fix — two parts

1. Classify method-rejection as transient

`isTransientRpcError` now also matches:

• `rpc method is not whitelisted` (Flashbots-specific)
• `method not found` / `method not supported` / `method does not exist` (generic)
• `-32601` (canonical JSON-RPC code)
• `403` (catches ethers' `server response 403` wrapper)

These are URL-level capability problems; trying another URL helps. Treating them as transient means the URL gets a 60s cooldown via `failedRpcs`.

2. All 14 EVM read handlers migrate to `withRpcFailover`

`eth_getBlockByNumber`, `eth_blockNumber`, `eth_getBalance`, `eth_getTransactionReceipt`, `eth_getTransactionByHash`, `web3_clientVersion`, `eth_call`, `eth_max{Priority,}FeePerGas`, `eth_estimateGas`, `eth_gasPrice`, `eth_feeHistory`, `eth_getCode`, `eth_getStorageAt`, `eth_getTransactionCount` — each now runs through the failover loop with classification, transport timeout, and per-attempt URL rotation.

Plus two cross-cutting reads also migrated:

• The fee-warning preflight bundle (4 reads in `Promise.all` on the sign path) wrapped as a single op so the four reads stay consistent across an RPC.
• The side-panel transfer build (nonce + estimateGas + getFeeData) — each fails over independently.

`getProvider()` remains only as the drop-check fallback when the SW restart loses the success-URL hint. All other paths share `failedRpcs` through `withRpcFailover` so failover decisions are consistent.

Files

• `chrome-extension/src/background/chains/ethereumHandler.ts` — `isTransientRpcError` expanded; 14 read handlers + 2 preflight bundles migrated.

Test plan

• Visit CowSwap on Ethereum mainnet — multicall reads no longer 403; quote loads.
• Visit Uniswap, do a multi-hop swap quote — `eth_call` reads complete.
• Trigger Permit2 simulation (Uniswap pre-quote) — `stateOverride` passthrough still works.
• Watch console: after the first Flashbots 403, subsequent reads should hit a different URL ([rpcFailover] log shows `transient failure, trying next:`).
• Wait 60s; Flashbots should re-enter the rotation. Browse to a non-call route (just balance) — Flashbots may or may not be picked first depending on candidate ordering.
• Existing flows (BSC switch, signMessage, broadcast failover) — no regression.

🤖 Generated with Claude Code