feat: add tenacity retry to Grow/LevelData API requests

[cbini]

Problem

The Grow API resource (src/teamster/libraries/level_data/grow/resources.py) has no retry logic on HTTP requests. When grow-api.leveldata.com times out, the step fails immediately and the entire run relies on Dagster's run-level auto-retry — restarting the full run instead of retrying just the failed request.

On 2026-04-07, kipptaf__grow__multi_partition_assets_schedule failed due to a connection timeout to grow-api.leveldata.com. The Dagster auto-retry succeeded 2 minutes later, but the entire run was re-executed rather than just the timed-out request.

Proposed Change

Add tenacity retry with exponential backoff + jitter to GrowResource._request(), matching the pattern already used in:

• ADP WFN (libraries/adp/workforce_now/api/resources.py:53-57) — @retry(stop=stop_after_attempt(5), wait=wait_exponential_jitter())
• Google Forms (libraries/google/forms/resources.py:35) — same pattern
• Extracts (libraries/extracts/assets.py:105-109) — same pattern

Retry on:

• requests.exceptions.ConnectionError (includes timeouts when timeout kwarg is set)
• requests.exceptions.Timeout
• HTTP 429 (rate limit) — respect Retry-After header if present
• HTTP 502, 503, 504 (transient server errors)

Scope

• GrowResource._request() in src/teamster/libraries/level_data/grow/resources.py
• Also consider GrowResource._get_access_token() (OAuth2 token fetch can also time out)
• Check OvergradResource — same level_data library, also has no retry logic and a TODO noting missing exponential backoff (libraries/overgrad/resources.py:32)

Out of Scope

Broader audit of all API resources for missing retry logic (separate issue).