Fix span attribute size estimation

Include attribute key length and realistic per-attribute overhead in span size estimation to avoid underestimating small attributes.

Co-Authored-By: GPT-5.3 Codex High <noreply@anthropic.com>

Author: cursoragent

sentry_sdk/_span_batcher.py

@@ -94,8 +94,8 @@ def _estimate_size(item: "StreamedSpan") -> int:
         # 210 is the rough size of the payload without attributes, and then we
         # estimate the attributes separately.
         estimate = 210
-        for value in item._attributes.values():
-            estimate += 7
+        for key, value in item._attributes.items():
+            estimate += 40 + len(key)
 
             if isinstance(value, str):
                 estimate += len(value)