Typed response models (TypedDict/dataclasses) for endpoint returns

[martinkersner]

Problem

Endpoints return raw dict/list (or DataFrames), so users index by guessed keys (res["data"][0]["p"]) with no autocomplete or type checking on outputs. PR #136 gave Literal typing to inputs; outputs have none.

Proposal

Add lightweight typed models per response (start with the high-traffic ones: candle, ticker, premium):

• TypedDict for the dict-return path (cheap, no runtime cost), or
• dataclasses if we want attribute access + validation.
• Keep pandas=True DataFrame path as-is.

Notes

• Could be generated from the same OpenAPI spec that feeds _endpoints.py (codegen upstream) to avoid drift.
• Incremental: type one endpoint family at a time.