arctl configure fix batch

[inFocus7]

Description

Motivation
A few issues with arctl configures output. To name a few:

1. We are overriding existing MCP server configs with our un/marshalling that was based expected narrow structs, deleting unrelated keys and only modifying our managed config (arctl).
   • e.g. In Cursor/Kiro existing stdio servers became {"url": ""} in Cursor/Kiro; VS Code lost top-level inputs/sandbox, Claude Code dropped headers/timeout/oauth from other servers.
2. (Minor) Naming inconsistency. Different configs had different mcp server naming (arctl vs ARCTL)
3. We have no way to configure a bearer token for the mcp bridge. We recently we've added support for vendor ext. to configure mcp bridge auth, so this is a good additions to that.

What Changed

1. Stopped marshalling to specific keys because each client configures a lot of unrelated keys that we must preserve. Now we do a raw marshal to preserve existing information.
2. Changed naming in some Clients from ARCTL to arctl with cleanup to avoid end-user changes needed.
3. Added new flag --token-env allowing direct-access config to MCP bridge if it requires auth.
   • This is different from authenticating via the OAuth flow in interactive clients.
4. Additional testing

Change Type

/kind feature
/kind fix

Changelog

Fix issue where `arctl configure` was removing unrelated keys from local config files, and support static MCP bridge token for `arctl configure` via `--token-env VAR_NAME`.

Additional Notes

Does not set up any auth or changes to the server. Simply focused on input (cmd) -> output (config file) translation which is verified via tests and local runs of arctl configure.

The largest unknown is Kiro and if it requires ${TOKEN} vs ${env:TOKEN} where there is an open issue stating the former documented version does not work. If anyone with Kiro tests locally and hits an issue, we likely just need to remove the env: prefix to the variable.