Deduplicate or replace built-in LSP servers when user config defines the same server id

[darkodus]

Description

When configuring AFT LSP servers, defining a custom server with the same id as a built-in server appears to create multiple matching server entries instead of replacing or clearly merging with the built-in one.

Example intent:

{
  "lsp": {
    "servers": {
      "clangd": {
        "args": ["--query-driver=/path/to/arm-none-eabi-*"],
        "root_markers": ["CMakeLists.txt", ".clangd", "compile_commands.json", ".git"]
      }
    }
  }
}

Expected behavior would be that the configured clangd entry replaces or predictably merges with the built-in clangd server.

Observed behavior was that diagnostics appeared to use multiple clangd entries for the same file, with one server using the default config and another using the custom config/root markers. The workaround was to disable the built-in server and register a differently named server, for example custom-clangd.

Use case

Embedded C/C++ projects often need to customize clangd with cross-compilation arguments such as --query-driver, custom root markers, or compile-database settings. It would be cleaner if users could customize the built-in clangd server directly while keeping the canonical server id.

Current workaround:

{
  "lsp": {
    "disabled": ["clangd"],
    "servers": {
      "custom-clangd": {
        "extensions": [".c", ".cc", ".cpp", ".cxx", ".h", ".hh", ".hpp", ".hxx"],
        "binary": "clangd",
        "args": ["--query-driver=/path/to/arm-none-eabi-*"],
        "root_markers": ["CMakeLists.txt", ".clangd", "compile_commands.json", ".git"]
      }
    }
  }
}

This works, but it feels like a workaround and makes global config project-specific. Ideally users could keep the server id as clangd and override or extend only the fields they need.