fix(config): disable options rejected by bee in light and ultra-light…

[martinconic]

Companion to ethersphere/bee#5596. Supersedes #580, whose changes this branch carries.

What changes

Profiles move from full-node to node-mode. Every bee-config in config.yaml, local.yaml, light-node.yaml, public-testnet.yaml, staging.yaml and testnet-bee-playground.yaml now sets node-mode (full, light or ultra-light) instead of full-node.

Light and ultra-light profiles stop inheriting options bee rejects. With node-mode set, bee rejects storage incentives outside full mode and a chequebook without swap. These profiles inherited both as true from their full-node parent:

Profile	mode	swap	chequebook	incentives
bee-local-light	light	true	true	false
bee-local-light-autotls	light	true	true	false
bee-local-ultralight-autotls	ultra-light	false	false	false
light-node	light	true	true	false

The ultra-light profile now really has no endpoint. bee-local-ultralight-autotls set blockchain-rpc-endpoint: to null, which inheritance treats as unset, so it inherited ws://geth-swap:8546 from bee-local-dns: on master that node has been running as a light node. It is now blockchain-rpc-endpoint: "", which clears the inherited value, so the node is a genuine ultra-light node and bee does not warn that the endpoint is ignored.

Bootnodes disable storage incentives. The bootnode profiles (bootnode, bootnode-local, bootnode-local-dns, bootnode-local-dns-autotls, bootnode-local-gc, geth-playground-bootnode) inherited storage-incentives-enable: true. Bee never runs the incentives agent on a bootnode and, with node-mode set, warns about it on every start; the profiles now set it to false. Nothing changes at runtime.

The local-dns cluster also runs one node in bee's legacy regime. A new bee-legacy group (1 node, profile bee-local-dns-legacy) overrides the inherited node-mode with "" and sets full-node: true, so bee infers the mode the way releases before node-mode did. It runs next to the 3 node-mode: full nodes, so bee#5596's CI covers both regimes in one cluster.

Orchestration. orchestration.Config gains NodeMode plus IsFullNode, IsLightNode and IsUltraLightNode, which read node-mode and fall back to full-node (and the RPC endpoint) when it is unset or empty. FullNodeNames, LightNodeNames, ShuffledFullNodeClients and ClosestFullNodeClient use them, and UltraLightNodeNames is added.

CI. golangci-lint bumped to v2.11.3 with the lint cache disabled.

Merge order

Bee releases up to and including v2.8.3 reject unknown config keys at startup (CheckUnknownParams), and they have no node-mode option. A node from any of those releases refuses to start with a .bee.yaml rendered from these profiles. This PR should therefore merge only after bee#5596 is merged and released, since the remote profiles (staging, public-testnet, testnet-bee-playground) deploy released images.

Testing

• TestNodeModeHelpers covers the helpers for explicit modes and the full-node fallback.
• TestLocalDNSLegacyNodeGroup loads config/local.yaml and checks that the bee-legacy node renders node-mode: "" with full-node: true, while bee-local-dns keeps node-mode: full.
• A bee binary built from bee#5596's branch, started with the rendered legacy profile, logs the legacy-inference warning with node_mode="full"; the bee-local-dns profile resolves to full without it.
• Every bee-config across all config/*.yaml files (19 profiles) was rendered by beekeeper and resolved by bee#5596's real config pipeline: all resolve without warnings except the intended migration warning on bee-local-dns-legacy. The only profile bee rejects is sepolia (full-node/node-mode: full without an RPC endpoint), unchanged from master, where bee fails it at chain init; its endpoint is expected to come from outside beekeeper.
• go vet, the pkg/config and pkg/orchestration tests, and make lint pass. bee#5596's Beekeeper workflow is pinned to this branch, so its run exercises the cluster end to end.