Tidy/Improve BaseClient Implementation

[louispt1]

Current State

• BaseClient (singleton) wraps an ETMSession.
• AsyncBatchRunner.batch_requests runs a list of loosely-typed request dicts concurrently with asyncio.gather and returns a list of ServiceResult.
• A sync wrapper (batch_requests_sync) uses run_coroutine_threadsafe on a private event loop.
• A free function make_batch_requests duplicates the sync path.
• Error handling is broad; request specs are untyped dicts; results only available after all complete.

Drawbacks

• Singleton pattern reduces flexibility (shared global state).
• Batch API is all-or-nothing: no streaming / incremental consumption.
• Uses private session loop attribute; brittle.
• Request specification is an unvalidated dict → easy to introduce mistakes.
• Helper make_batch_requests feels out of place.
• No concurrency limiting, progress callbacks, or structured retry/parse strategy.
• Error classification is coarse; no partial yield or cancellation friendliness.
• Hard to extend (no hooks / observers).

Ideal State

• Instantiable (optionally cacheable) client with context manager support (sync + async).
• Typed RequestSpec (dataclass) defining method, url, kwargs, optional id.
• Streaming batch API (e.g. async iterator using asyncio.as_completed) plus convenience collector.
• Optional concurrency limit, progress/observer callbacks, and retry/backoff hooks.
• Clear error classification (auth/client/server/parse) with structured data in ServiceResult.
• Sync adapter that doesn’t rely on private loop internals (fails fast if misused inside running loop).
• Integrated logging/tracing; optional response parser injection.
• Deprecated/removal of the free helper in favor of a client.batch(...) method.