docs: fix e2e test port ranges in docs/testing.md

[erkancamli]

The port ranges in the E2E "Prerequisites" section do not match what the test bins bind, and the first revision of this PR got the correction half wrong. This revision lists every port and states the mapping the numbers come from.

Reth's --instance N moves four different ports, in three different directions. Each instance is spawned with .instance(x + 1) (node/src/bin/stake_and_checkpoint.rs:134, and the same in the other e2e bins). Reth's adjust_instance_ports then applies:

auth_port += instance * 100 - 100
http_port -= instance - 1
ws_port   += instance * 2 - 2

and NetworkArgs::adjust_instance_ports shifts the discovery port by instance - 1. Instances 1 through 6 are used (four genesis nodes, a fifth node in three bins, and observer at OBSERVER_SLOT = NUM_NODES + 1 = 5, node/src/bin/observer.rs:62,251), so:

	instances 1..6
HTTP RPC	8545, 8544, 8543, 8542, 8541, 8540
WS	8546, 8548, 8550, 8552, 8554, 8556
auth RPC	8551, 8651, 8751, 8851, 8951, 9051
discovery	30303 .. 30308

HTTP counts down, as docs/running-local-network.md:35 and the header of node/src/bin/testnet.rs say. WS and auth count up, which neither doc mentions. types/src/reth.rs always passes --ws (line 515) and spawn() blocks until reth reports both RPC WS server started and RPC auth server started, so those ports are provably bound on every e2e run.

Each bin also passes --metrics 0.0.0.0:{9001 + node} and --enclave.endpoint-port 1744{node} explicitly, so 9001-9006 and 17440-17445 are bound too.

The Summit and P2P ranges were sized for the four genesis nodes. get_node_flags derives 3030 + node * 10 (RPC), 3031 + node * 10 (admin RPC) and 26600 + node * 10 (P2P). With the fifth node and the observer these reach 3080/3081 and 26650.

Following the current text, a reader frees ports nothing uses and leaves ports the tests need occupied. The run then fails as a node that never reaches its stop height rather than as a clear bind error.

Documentation only, no code change. Same class of fix as #441, which corrected this in docs/running-local-network.md but did not touch docs/testing.md. Filed against fa436b4.

---

Correcting the record on my own first revision: it said "so 8546-8548 are never bound while 8542-8544 are". That is wrong. 8546 and 8548 are the WS ports of instances 1 and 2, and types/src/reth.rs waits for them before returning. Only 8547 is unbound. The narrowed 8540-8545 range it proposed would have caused exactly the failure this PR is meant to prevent. Fixed in ab0f360.