feat: phase 2 — am lab (benchmark N agents) + plugins d'événements (contrat JSON)

- #62: am lab --agents a,b --task <f> — séquentiel ou --parallel, capture
  durée/exit/sortie/coût estimé, rapport comparatif tableau + --json stable,
  tâches versionnables dans <state>/lab/ (--list), rejouables à l'identique
- #75: plugins/ sous state_dir, déclenchés sur on_install/on_start/on_stop/
  on_update (App::emit), contrat JSON stdin/stdout, timeout configurable,
  échec non bloquant, garde anti-récursion AM_PLUGINS_RUNNING, am plugins
  + am plugins --test <name> (CI), exemples notify + ci-webhook
- intégration REPL (parse, complétion, bannière, is_am_command), help, tips,
  man pages (am-lab.1, am-plugins.1), README, ROADMAP (26/27), v0.5.4
- 334 tests verts (309 + 25)

closes #62
closes #75

man/am-lab.1

@@ -0,0 +1,28 @@
+.ie \n(.g .ds Aq \(aq
+.el .ds Aq '
+.TH am-lab 1  "lab " 
+.SH NAME
+lab \- Benchmark: run the same task on several agents and compare (issue #62)
+.SH SYNOPSIS
+\fBlab\fR [\fB\-\-agents\fR] [\fB\-\-task\fR] [\fB\-\-parallel\fR] [\fB\-\-timeout\fR] [\fB\-\-list\fR] [\fB\-h\fR|\fB\-\-help\fR] 
+.SH DESCRIPTION
+Benchmark: run the same task on several agents and compare (issue #62)
+.SH OPTIONS
+.TP
+\fB\-\-agents\fR \fI<AGENTS>\fR
+Comma\-separated agent names (or aliases) to benchmark
+.TP
+\fB\-\-task\fR \fI<FILE>\fR
+Task file (a bare name resolves under state_dir/lab/, .yaml appended)
+.TP
+\fB\-\-parallel\fR
+Run every agent concurrently instead of one after another
+.TP
+\fB\-\-timeout\fR \fI<SECS>\fR
+Override the task timeout in seconds
+.TP
+\fB\-\-list\fR
+List the task files of the lab directory
+.TP
+\fB\-h\fR, \fB\-\-help\fR
+Print help