docs: sync specs to implemented state, remove max_startup_latency_seconds

[mohammadp1001]

Summary

• Removes the max_startup_latency_seconds config field entirely (it only gated a warning during candle-boundary alignment; never blocked or skipped a session) from AlphorynConfig, Scheduler.wait_for_candle_close, config.json.example, and tests.
• Brings specs/001-etf-paper-trading-agent/ back in line with the actual implemented codebase, which had drifted in several places:
  • Finishes the ETF → ticker terminology rename in contracts/agents.md, contracts/report-context.md, and research.md (PR Generalise ETF terminology to ticker/asset throughout codebase #99 covered the codebase but missed these three spec docs — checklists/requirements.md had a stale claim that this was already done).
  • Corrects data-model.md's Session table and SignalSnapshot to match the real generalized N-ticker schema (ticker_decisions JSON blob, signals: dict[str, AssetSignals]) instead of stale fixed ticker1_*/ticker2_* fields.
  • Corrects contracts/agents.md to the real AssetDecision/SessionDecision dataclasses and execution/telemetry behavior.
  • Corrects contracts/cli.md and quickstart.md to the real --tickers flag (was documented as --etf1/--etf2).
  • Corrects contracts/config-schema.md to the real tickers: list[str] config shape.
  • Corrects research.md's telemetry event table to match what's actually emitted, flagging known implementation gaps (execution/agent.py has no TelemetryLogger wired in; EVENT_TYPES is missing EVALUATION_FAILED).
  • Corrects plan.md skill paths to the real hyphenated directory names.
• tasks.md was deliberately left untouched — it's a completed historical task log with [x] checkboxes, not a living spec.

Test plan

• ruff check . — all checks passed
• pytest --cov=alphoryn — 444 passed, 100% coverage