fix: correct context pruning order and vector alignment

- Prune context BEFORE processing (LSTM forget gate pattern)
  so LLM only sees relevant context, instead of pruning after the fact
- Fix ensureTrained() to recompute all event vectors after retraining
  vectorizer, fixing feature-space mismatch between stored vectors and
  query vector that made relevance scoring effectively random
- Add read lock to knowledge BuildTree() (data race fix)
- Log writeIndex() errors instead of discarding them
- Fix TOCTOU race in document ContextToDoc() dedup
- Fix healthcheck timing (measure elapsed before cleanup)
- Refactor waiter CLI into separate files (state, conn, config, editor,
  history, builtin) for maintainability
- Add CLI plugin API key authentication

cmd/waiter/history.go

@@ -0,0 +1,58 @@
+package main
+
+import (
+	"os"
+	"strings"
+	"sync"
+)
+
+type History struct {
+	path  string
+	lines []string
+	max   int
+	mu    sync.Mutex
+}
+
+func newHistory(path string, max int) History {
+	return History{path: path, max: max}
+}
+
+func (h *History) load() {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	data, err := os.ReadFile(h.path)
+	if err != nil {
+		return
+	}
+	h.lines = strings.Split(strings.TrimSpace(string(data)), "\n")
+	if len(h.lines) > h.max {
+		h.lines = h.lines[len(h.lines)-h.max:]
+	}
+}
+
+func (h *History) save() {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	data := strings.Join(h.lines, "\n") + "\n"
+	os.WriteFile(h.path, []byte(data), 0644)
+}
+
+func (h *History) add(line string) {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	if len(h.lines) > 0 && h.lines[len(h.lines)-1] == line {
+		return
+	}
+	h.lines = append(h.lines, line)
+	if len(h.lines) > h.max {
+		h.lines = h.lines[len(h.lines)-h.max:]
+	}
+}
+
+func (h *History) all() []string {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	r := make([]string, len(h.lines))
+	copy(r, h.lines)
+	return r
+}