fix: strip carriage returns and newlines from base64 attachment data

Fixes #494 - Windows attachment corruption bug

## Root Cause

Windows text-mode HTTP response processing converts embedded LF to CRLF in
base64 strings, resulting in double carriage returns (0d0d0a) in the decoded
binary output. Python's base64.urlsafe_b64decode() is strict and passes this
corrupted input through. Node.js implementations silently strip the corruption.

## Solution

Strip CR and LF from the base64 string before decoding. Per RFC 4648 Section 3.1,
base64 decoders should ignore embedded whitespace.

## Impact

- Fixes PNG/PDF corruption on Windows for large attachments (>10KB)
- No impact on Unix/Linux/Mac
- Safe across all platforms
- Small attachments unaffected (no embedded line breaks in base64)

## Testing

Comprehensive unit tests included covering Unix LF, Windows CRLF, corrupted
double-CR, and large PDF-like base64 strings.

Author: mickey-mikey

gmail/gmail_tools.py

@@ -887,7 +887,11 @@ async def get_gmail_attachment_content(
     # Format response with attachment data
     size_bytes = attachment.get("size", 0)
     size_kb = size_bytes / 1024 if size_bytes else 0
-    base64_data = attachment.get("data", "")
+    # Strip CR and LF from base64 string (Windows text-mode corruption fix)
+    # RFC 4648 allows whitespace in base64; Python's urlsafe_b64decode is strict
+    # Node.js implementations (e.g., GongRzhe) don't have this issue
+    base64_data = attachment.get("data", "").replace("", "").replace("
+", "")
 
     # Check if we're in stateless mode (can't save files)
     from auth.oauth_config import is_stateless_mode