feat: state machine for scrape task producer state mgmt

[extreme4all]

replaces imperative FetchParams mutation with generic state machine. Good separation: sm.py (generic engine), structs.py (enums/context), states.py (transition declarations), core.py (orchestration). ~120 lines cut from core.

Summary

Refactor scrape task producer state management from mutable FetchParams to an explicit state machine pattern.

Changes

• Add generic state machine (sm.py): StateMachine[S, E, C] with enum states/events, context object, decorator-based transition registration.
• Extract domain types (structs.py): ScrapeState enum (NORMAL → POSSIBLE_BAN → CONFIRMED_BAN → DONE), ScrapeEvent enum (FETCH_MORE, REDUCE_DAYS, NEXT_STEP, NEW_DAY), ScraperCtx dataclass replacing FetchParams.
• Declare transitions (states.py): All state transitions and side effects via @scraper_sm.transition decorators.
• Simplify core (core.py): Replace determine_fetch_params (mutation) with determine_event (pure) + scraper_sm.handle (transition). Remove run_async/run wrappers, inline main() setup.
• Rewrite tests: Test state transitions and event determination independently.

State flow

NORMAL → POSSIBLE_BAN → CONFIRMED_BAN → DONE
  ↑                                    |
  └──────────── NEW_DAY ───────────────┘

Each state narrows the date window (REDUCE_DAYS) until exhausted, then advances (NEXT_STEP). FETCH_MORE pages via cursor.

Notes

fetch_more uses three decorators (one per state → self) as a workaround for missing self-transition concept in the generic SM. Consider adding transition_to_self or allowing to_state=None to mean "stay."