v2.0.0 — Pure glue architecture: zero copied code, pluggable channels

BREAKING: Complete architectural rewrite.

Before: Copied x-reader's fetcher code into readers/ (1205 lines of borrowed code)
After: Pluggable channel system where each channel is a thin wrapper (~50 lines)
       around the best external tool for that platform. Zero copied code.

Architecture:
- channels/base.py — Universal Channel interface (read, search, check)
- channels/web.py — Jina Reader API (swappable)
- channels/github.py — GitHub API (swappable)
- channels/twitter.py — birdx + Jina fallback (swappable)
- channels/youtube.py — yt-dlp (swappable)
- channels/reddit.py — Reddit JSON API + proxy (swappable)
- channels/rss.py — feedparser (swappable)
- channels/bilibili.py — Bilibili API (swappable)
- channels/exa_search.py — Exa semantic search (swappable)

Key design: every backend can be swapped by changing ONE file.
YouTube dies? Change youtube.py. Exa sucks? Swap exa_search.py for Tavily.
Nothing else changes.

Removed: reader.py, schema.py, readers/, search/, utils/ (all x-reader code)
Tests: 36/36 passing

agent_eyes/channels/bilibili.py

@@ -0,0 +1,77 @@
+# -*- coding: utf-8 -*-
+"""Bilibili — via public API (free, no config needed).
+
+Backend: Bilibili public API
+Swap to: any Bilibili access method
+"""
+
+import requests
+from urllib.parse import urlparse, parse_qs
+from .base import Channel, ReadResult
+
+
+class BilibiliChannel(Channel):
+    name = "bilibili"
+    description = "Bilibili video info and subtitles"
+    backends = ["Bilibili API"]
+    tier = 0
+
+    def can_handle(self, url: str) -> bool:
+        domain = urlparse(url).netloc.lower()
+        return "bilibili.com" in domain or "b23.tv" in domain
+
+    async def read(self, url: str, config=None) -> ReadResult:
+        # Extract BV id from URL
+        path = urlparse(url).path
+        bv_id = ""
+        for part in path.split("/"):
+            if part.startswith("BV"):
+                bv_id = part
+                break
+
+        if not bv_id:
+            # Fallback to Jina Reader
+            from agent_eyes.channels.web import WebChannel
+            return await WebChannel().read(url, config)
+
+        # Get video info
+        resp = requests.get(
+            "https://api.bilibili.com/x/web-interface/view",
+            params={"bvid": bv_id},
+            headers={"User-Agent": "Mozilla/5.0"},
+            timeout=15,
+        )
+        resp.raise_for_status()
+        data = resp.json().get("data", {})
+
+        title = data.get("title", "")
+        desc = data.get("desc", "")
+        author = data.get("owner", {}).get("name", "")
+
+        # Try to get subtitles
+        subtitle_text = ""
+        subtitle_list = data.get("subtitle", {}).get("list", [])
+        if subtitle_list:
+            sub_url = subtitle_list[0].get("subtitle_url", "")
+            if sub_url:
+                if sub_url.startswith("//"):
+                    sub_url = "https:" + sub_url
+                sr = requests.get(sub_url, timeout=10)
+                if sr.ok:
+                    sub_data = sr.json()
+                    lines = [item.get("content", "") for item in sub_data.get("body", [])]
+                    subtitle_text = "\n".join(lines)
+
+        content = desc
+        if subtitle_text:
+            content += f"\n\n## Transcript\n{subtitle_text}"
+
+        return ReadResult(
+            title=title,
+            content=content,
+            url=url,
+            author=author,
+            platform="bilibili",
+            extra={"view": data.get("stat", {}).get("view", 0),
+                   "like": data.get("stat", {}).get("like", 0)},
+        )