fix: don't retry on QueueingError (Veo/Lyria async queueing is not a transient error)

[chiuweilun1107]

Problem

When generate_content encounters Stream suspended (queueing=True), the server has accepted the request and is processing it asynchronously (e.g. Veo video rendering, Lyria music generation). The current code raises APIError at this point, which the @running(retry=5) decorator treats as a transient error and retries.

Each retry sends a new request, creating a new server-side job (new conversation) and burning an additional daily quota slot.

Empirical evidence

Tested with Veo 3 video generation prompts:

• Within 45 seconds, the decorator fired 4 retries, creating 4 independent Veo conversations visible in the Gemini web UI
• Each consumed a separate daily quota slot
• The backoff delays (5s/10s/15s/20s/25s) don't help because Stream suspended fires within ~12s of each attempt — faster than the first backoff

Fix

Introduce QueueingError(GeminiError) — a new exception that inherits from GeminiError instead of APIError.

When is_queueing=True at the point of stream suspension:

• Raise QueueingError instead of APIError
• The @running decorator only retries APIError, so QueueingError bubbles up immediately with zero retries
• Callers can catch QueueingError and switch to a poll-based flow (list_chats() + read_chat()) to retrieve the result once the server finishes rendering

When is_queueing=False (transient connection issues, cookie drift, etc.):

• Still raises APIError as before — existing retry behaviour is preserved

Changes

File	Change
exceptions.py	Add QueueingError(GeminiError)
client.py	Import QueueingError; raise it instead of APIError when is_queueing=True in the stream-suspended branch

No changes to decorators.py — the fix works purely through the exception hierarchy.

Backward compatibility

• Code that catches APIError will not catch QueueingError — this is intentional (the retry was harmful, not helpful)
• Code that catches GeminiError will catch QueueingError (since it's a subclass)
• Code that catches Exception is unaffected

Usage example (after this fix)

from gemini_webapi.exceptions import QueueingError

try:
    response = await client.generate_content("Generate a video of a cat")
except QueueingError:
    # Server accepted the request and is rendering asynchronously.
    # Poll list_chats() for the new conversation, then read_chat(cid)
    # to retrieve the generated video URL.
    new_chats = await client.list_chats()
    # ... poll and download

🤖 Generated with Claude Code

[luuquangvu]

Gemini only actually receives the request and begins processing it when a CID appears, and when a CID is present, the system waits until a result is available. Therefore, the analysis in this PR is incorrect.