refactor(lsp): cache analysis results and debounce diagnostics

This change adds cached and incremental analysis for the LSP
server to improve responsiveness. The client now debounces
diagnostic updates to avoid excessive recomputation. The
server introduces per-document line caches and various
caches for completions, inlay hints, and metadata to
support faster, incremental updates.

- Debounce diagnostics on text changes to reduce noise.
- Add caches for completions, inlay hints, and metadata.
- Introduce incremental analysis with per-document line caches.

server/src/tools/completion_items.py

@@ -1,8 +1,9 @@
-from tclint.syntax_tree import Visitor, Command, BareWord, List
 import lsprotocol.types as lsp
 from common.load_data import standard_items
+from tclint.syntax_tree import BareWord, Command, List, Visitor
 
 BUILTIN_VAR_LABELS = {ci.label for ci in standard_items.nx_variables}
+BUILTIN_PROC_LABELS = {ci.label for ci in standard_items.nx_procs}
 
 
 class CompletionItems:
@@ -22,6 +23,7 @@ class CompletionCollector(Visitor):
     def __init__(self):
         super().__init__()
         self._custom_functions: list[lsp.CompletionItem] = []
+        self._custom_function_keys: set[tuple[str, lsp.CompletionItemKind | None]] = set()
         self._proc_signatures = {}
 
     @property
@@ -34,8 +36,11 @@ class CompletionCollector(Visitor):
 
     def _append_unique(self, item: lsp.CompletionItem):
         # Avoid duplicate labels within the same file scan
-        if not any(ci.label == item.label for ci in self._custom_functions):
-            self._custom_functions.append(item)
+        key = (item.label, item.kind)
+        if key in self._custom_function_keys:
+            return
+        self._custom_function_keys.add(key)
+        self._custom_functions.append(item)
 
     def visit_command(self, command: Command):
         routine = command.routine
@@ -46,7 +51,7 @@ class CompletionCollector(Visitor):
             if not getattr(first_arg, "value", None):
                 return
 
-            if any(item.label == first_arg.value for item in standard_items.nx_procs):
+            if first_arg.value in BUILTIN_PROC_LABELS:
                 return
 
             # Record proc name as a completion item