[Feature] create_and_wait helpers for Flows generations and template runs

[brandonrule11]

FLO-350: quickstarts should get a Flows result in one call, without a hand-written poll loop or webhook plumbing (like OpenAI's create_and_poll). _should_retry only covers 408/409/429/5xx, so an in-progress 200 returns straight away. JS counterpart: elevenlabs/elevenlabs-js#469.

Usage

gen = client.flows.image.create_and_wait(request=ImageGenerationRequest_BytedanceSeedream5Lite(prompt="a corgi"))
run = client.flows.templates.runs.create_and_wait("template_id", inputs={}, timeout=600)

Design

• New src/elevenlabs/flows_custom.py, added to .fernignore. Subclasses of the generated flows, templates, runs, image, video and TTS clients, sync and async. client.py sets self._flows to the wrapper, the same pattern as music/STT/speech engine.
• Create and every GET go through with_raw_response, so the create response's Retry-After (FLO-349 sets it on creates too) sets the first wait.
• Delay uses the existing core.http_client._parse_retry_after and _add_positive_jitter: max(Retry-After or min, min) × (1 + 0–20%). A missing, unparseable or 0 header falls back to min_poll_interval (default 1s). Transport retries still handle 429s.
• completed/failed are terminal. A failed result is returned, not raised.
• With timeout (seconds), the last sleep is cut short at the deadline and one final GET runs. If the work still isn't finished, it raises FlowsWaitTimeoutError (id, template_id, last_response). With no timeout set, it waits indefinitely, like OpenAI.
• The docstrings point production users to webhooks.

Notes

• FlowsWaitTimeoutError is a plain Exception, not ApiError, because no HTTP status applies. Import it from elevenlabs.flows_custom, since the package __init__ is generated.
• I didn't run ruff format on the test file: ruff rewrites it to parenthesized context managers, which aren't valid on Python 3.8 (python = "^3.8").
• Generation docstrings don't name a webhook event. Only flows_template_run appears in the API definition.

Testing

• tests/test_flows_create_and_wait.py: covers Retry-After pacing and jitter, the min fallback, failed generations returned rather than raised, timeout errors carrying the id (and template_id for runs), the no-poll path when create returns a terminal status, and the async image and run paths. Uses httpx MockTransport with sleeps patched.
• pytest tests/test_flows_create_and_wait.py: 10/10 pass. mypy on the new and changed files is clean.
• Not run against the live API. FLO-349 hasn't shipped, so until it does the helper polls at the minimum interval.

🤖 Generated with Claude Code

[cursor]

PR Summary

Low Risk
SDK-only convenience layer over existing Flows APIs; polling and error semantics are covered by unit tests with no auth or billing changes.

Overview
Adds client.flows on sync and async ElevenLabs clients, backed by a new flows_custom.py wrapper (listed in .fernignore) that subclasses the generated Flows clients.

The main capability is create_and_wait on flows image, video, text_to_speech, and templates.runs: it creates the job, then polls get until status is completed or failed, honoring Retry-After (with jitter and a minimum interval) from create and poll responses. Optional timeout raises FlowsWaitTimeoutError with ids so callers can resume; failed results are returned, not raised. Docstrings steer production use toward webhooks.

tests/test_flows_create_and_wait.py exercises polling, jitter/min fallback, timeouts, terminal-on-create, sync/async paths, and failed generations via httpx MockTransport.

^{Reviewed by Cursor Bugbot for commit 3782630. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit dd0db64. Configure here.}