api: invoke_contract fails fast when there's no node connection (#1695)

[vsnation]

Fix: invoke_contract answers immediately with -32019 "No connection to the node" when wallet-api has no node connection, instead of never answering and blocking later calls.

Why: the shader only starts once the wallet is synced (Wallet::DoInSyncedWallet). Without a node that never happens, so the request never completes. Later calls then queue behind it (default API), or fail with "Previous shader call is still in progress" (v6.0). Neither handler (v6.0, v6.1) checked the connection.

How to see the bug (master, no node needed)

beam-wallet restore --pass=p --wallet_path=w.db --seed_phrase="abandon;abandon;abandon;abandon;abandon;abandon;abandon;abandon;abandon;abandon;abandon;about;"
wallet-api --pass=p --wallet_path=w.db -n 127.0.0.1:1 -p 10000 --use_http=1 &      # no node on :1
curl --max-time 8 http://127.0.0.1:10000/api/wallet -d '{"jsonrpc":"2.0","id":1,"method":"invoke_contract","params":{"contract_file":"bvm/Shaders/vault/app.wasm","args":"role=manager,action=view","create_tx":false}}'
# master: no response (hangs); a second identical call hangs too

Improvement

	master	this PR
call 1	no response (8 s timeout)	{"error":{"code":-32019,"data":"No connection to the node",…}} immediately
call 2	no response	same error, immediately (not blocked)
with a reachable node	works	unchanged

How: V6Api keeps the nodeNetwork and checks it in both invoke_contract handlers before anything is queued, so no busy state is left behind. Without a nodeNetwork (e.g. wasm), nothing changes. It reuses ContractError, so there are no new error codes.

Testing: the table above (master vs this branch). New TestInvokeContractNoNodeConnection in wallet_test (an offline wallet with a real ShadersManager: each call must throw ContractError and leave the manager not busy). Without the fix it fails 2 checks; with it, it passes. Fork CI (running): https://github.com/vsnation/beam/actions/runs/37479393148

Fixes #1695

[vsnation]

Closing in favour of #2142, which fixes #1695 more completely (it fails calls only after a reconnect really fails, and releases queued calls too). Tested side by side, see my comment there.