fix(core): ensure compaction is more reliable, add reserve token buffer to ensure that input window has enough room to compact (#12924)

Co-authored-by: James Lal <james@littlebearlabs.io>

packages/web/src/content/docs/config.mdx

@@ -490,13 +490,15 @@ You can control context compaction behavior through the `compaction` option.
   "$schema": "https://opencode.ai/config.json",
   "compaction": {
     "auto": true,
-    "prune": true
+    "prune": true,
+    "reserved": 10000
   }
 }
 ```
 
 - `auto` - Automatically compact the session when context is full (default: `true`).
 - `prune` - Remove old tool outputs to save tokens (default: `true`).
+- `reserved` - Token buffer for compaction. Leaves enough window to avoid overflow during compaction
 
 ---