fix: address review findings (C1/C3/H1/H2/H5 + M1)

Tackles the critical and high-impact findings from the in-PR review
of commit 7f01be5, plus one consistency fix (M1). All changes are
scoped to the same review branch (PR #12); no behavior changes
elsewhere.

Critical fixes
* C1 -- plugin/22_languages.lua: now does `local Config = require('config')`
  at the top, matching every other `plugin/*.lua` file. Previously the
  file referenced `Config` as a global, depending on `_G.Config` having
  been initialized by `init.lua` before this file loaded. Fragile.
* C2 -- plugin/04_treesitter.lua: removed the hard `<CR>` ->
  `smart_send.send_repl` mapping. Override `<CR>` from a per-buffer
  `ftplugin/<lang>.lua` if you want enter-to-send behavior. R.nvim's
  `<Plug>RDSendLine` (wired by `ftplugin/r.lua`) remains the default
  for R files and is no longer silently clobbered.

High-impact fixes
* C3 -- plugin/27_image.lua: `image.setup({ backend = "kitty" })` ->
  `backend = "auto"`. The previous value silently failed on every
  terminal that is not Kitty. `"auto"` delegates detection to image.nvim.
* H1 -- plugin/10_keymap.lua: `_G.Config = Config` removed from this
  file; `init.lua:2` remains the single source. Avoids drift between
  two aliasing sites.
* H2 -- plugin/25_lsp.lua: `texlab = { single_file_support = true }`,
  so single-file `.tex` buffers attach the LSP without lspconfig's
  sometimes-brittle root_dir heuristic.
* H5 -- plugin/04_treesitter.lua: also drops `<S-CR>` from
  `M.setup_keybindings` on the same principle as C2. `<S-CR>` was a
  hard implicit override in both normal and insert mode, where it
  collided with snippet and transient-state plugins.

Consistency fixes
* M1 -- plugin/01_lib.lua: `print(line)` in `Config.execute_lua_line`
  switched to `vim.notify(line, vim.log.levels.INFO)`, matching the
  print->notify cleanup in `plugin/04_treesitter.lua` from 7f01be5.

Files: 6 modified. +21 / -11.

Local verification
1. `luac -p plugin/{22_languages,27_image,10_keymap,25_lsp,04_treesitter,01_lib}.lua`
2. `nvim --headless -u NONE -l tests/init.lua`
3. Open a R / quarto / tex buffer; verify `<CR>` is no longer hijacked
   by `smart_send` and behaves like the filetype default.

For reviewers
* The text-object configuration in plugin/04_treesitter.lua's
  textobjects block is unchanged. H4 (`@assignment.*` queries may not be
  defined for R) is left for a follow-up with verification.
* yamlls GitHub-rawURL schema dependency (H3) is left intentionally --
  vendoring the schemas is a separate decision.
* Tests (T1-T6 from the review) are deferred; no test infrastructure
  exists beyond `tests/init.lua`.

plugin/04_treesitter.lua

@@ -121,11 +121,17 @@ function M.setup_keybindings(global_nodes)
   vim.keymap.set('n', '<localleader>a', function() smart_send.send_repl(current_global_nodes) end,
     { noremap = true, silent = true, desc = "Send node to REPL", buffer = true })
 
-  vim.keymap.set({ 'n', 'i' }, '<S-CR>', function() smart_send.send_repl(current_global_nodes) end,
-    { noremap = true, silent = true, desc = "Send node to REPL", buffer = true })
-
-  vim.keymap.set('n', '<CR>', function() smart_send.send_repl(current_global_nodes) end,
-    { noremap = true, silent = true, desc = "Send node to REPL", buffer = true })
+  -- Both `<CR>` and `<S-CR>` were removed from `M.setup_keybindings`. They
+  -- were hard, implicit overrides that clobbered Vim/filetype defaults and
+  -- the user's snippet + insert-mode workflows (see C2 + H5 in the PR
+  -- review). To opt back in for a specific filetype, override per-buffer
+  -- from a `ftplugin/<lang>.lua`:
+  --
+  --   -- e.g. ftplugin/r.lua or ftplugin/quarto.lua
+  --   vim.keymap.set('n', '<CR>', function()
+  --     require('config').treesitter_helpers.setup_keybindings(global_nodes)
+  --     require('nix_smart_send').send_repl(global_nodes)
+  --   end, { buffer = true, desc = 'Send node to REPL' })
 
   vim.keymap.set('n', '<localleader>n',
     function() current_global_nodes = M.add_global_node(current_global_nodes) end,