fix: reuse the stored provider for incremental acquisition #95

Reproduction

On main55b86e250fd6279222876c5a1e4c20ae1469d7fa, write a real SyntheticProvider SPY daily frame to Hive/Flat with provider='synthetic' metadata, then DataManager(storage=storage).update('SPY', fill_gaps=False) without an explicit provider. It fails with No provider found for symbol SPY instead of using the recorded provider. The update reads existing_provider but passes the original None to both history-limit selection and fetch_raw. The stored provider is only reused when constructing the final metadata, after fetching already failed or used another route.

Expected behavior

Explicit provider overrides the recorded provider; otherwise reuse a non-empty recorded adapter for history limits and actual fetching. Legacy provider='auto' remains a routing sentinel rather than an adapter name. Invalid stored provider types still fail atomically when there is no override. Do not silently switch data sources or claim a provider in metadata that was not used for acquisition.

Use the actual offline SyntheticProvider through the public facade and storage backends, not a mocked fetch that ignores the chosen provider. Verify failing-base/passing-fix and full hosted gates. This is separate from calendar-selection PRs92/94. No release/deploy or heavy local suite during the workstation timing reservation. AI-assisted repair under maintainer direction.