feat(examples/nvim): add nvim-lsp config based option

examples/nvim.md

@@ -2,14 +2,61 @@
 
 ## Register the language server with neovim
 
-TODO: contribute some defaults to the nvim-lspconfig repo and link to that instead.
+You have 2 options, until a configuration is merged in
+[nvim-lspconfig](https://github.com/nvim-lspconfig/nvim-lspconfig), you can either use [my
+fork](https://github.com/Robzz/nvim-lspconfig) (recommended), or the raw nvim LSP API if you don't want to trust some
+random fork.
 
-The following code will register the language server with Neovim, and naively attach it to every buffer. `lsp-ai` is
-configured with the `llama_cpp` backend to run the CodeGemma v1.1 model.
+Both configurations configure `lsp-ai` to use the `llama_cpp` backend runnning an 8-bit quant of the CodeGemma v1.1
+model using FIM, fully offloaded to GPU.
 
 Note: the model configuration is provided as an example/starting point only and I do not vouch for the quality of the
 generations. Adjust to taste.
 
+### Using nvim-lspconfig
+
+Add the following snippet to your LSP configuration:
+
+```lua
+lspconfig.lsp_ai.setup {
+  -- Uncomment the following line if using nvim-cmp with the LSP source
+  -- capabilities = require('cmp_nvim_lsp').default_capabilities()
+
+  cmd_env = {
+    -- Add any environment variables you require here, e.g. for CUDA device selection
+    -- CUDA_VISIBLE_DEVICES = "1",
+  },
+  init_options = {
+    models = {
+      model1 = {
+        type = "llama_cpp",
+        repository = "mmnga/codegemma-1.1-2b-gguf",
+        name = "codegemma-1.1-2b-Q8_0.gguf",
+        n_ctx = 2048,
+        n_gpu_layers = 999
+      }
+    },
+    completion = {
+      model = "model1",
+      parameters = {
+        fim = {
+          start = "<|fim_prefix|>",
+          middle = "<|fim_suffix|>",
+          ["end"] = "<|fim_middle|>"
+        },
+        max_context = 2000,
+        max_new_tokens = 32
+      }
+    }
+  },
+}
+```
+
+### Using the nvim LSP API
+
+This configuration is not recommended for serious use as it does not properly manage the LSP server lifecycle and simply
+registers it with every buffer, even the ones you might not want, like terminal buffers.
+
 ```lua
 local lsp_ai_config = {
   -- Uncomment the following line if you use nvim-cmp with the cmp_nvim_lsp source.
@@ -19,9 +66,9 @@ local lsp_ai_config = {
     -- Add required environment variables here, e.g. for CUDA device selection.
     -- CUDA_VISIBLE_DEVICES = "1"
   },
-  root_dir = vim.loop.cwd(),
+  root_dir = nil,
   init_options = {
-    -- cmp-ai configuration goes here.
+    -- lsp-ai configuration goes here.
     memory = {
       file_store = {}
     },
@@ -65,7 +112,7 @@ For a copilot-like ghost-text experience, here is an example configuration using
 [cmp-nvim-lsp](https://github.com/hrsh7th/cmp-nvim-lsp) source. This is **not a full configuration**, please refer to
 the nvim-cmp documentation for a full starter config without ghost text if you need one.
 
-This configuration enables ghost-text in nvim-cmp, and registers a custom comparator that puts `cmp-ai` suggestions
+This configuration enables ghost-text in nvim-cmp, and registers a custom comparator that puts `lsp-ai` suggestions
 at the top so that they're the ones being drawn with ghost text.
 
 ```lua