feat: connect LLM to tri /chat — FluentChatEngine replaces pattern matcher

tri /chat now uses FluentChatEngine (symbolic + TinyLlama GGUF fallback)
instead of the pure pattern matcher (IglaLocalChat). This gives tri /chat
the same LLM intelligence as the fluent binary.

Changes:
- build.zig: vibeec_chat module root → igla_fluent_cli.zig
- igla_fluent_cli.zig: re-export IglaLocalChat types for backward compat
- tri/main.zig: CLIState uses FluentChatEngine, chat() instead of respond()

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: gHashTag

build.zig

@@ -339,7 +339,7 @@ pub fn build(b: *std.Build) void {
         .optimize = optimize,
     });
     const vibeec_chat = b.createModule(.{
-        .root_source_file = b.path("src/vibeec/igla_local_chat.zig"),
+        .root_source_file = b.path("src/vibeec/igla_fluent_cli.zig"),
         .target = target,
         .optimize = optimize,
     });