Add provider-neutral historical-to-realtime streaming

[luisleo526]

Summary

• add a continuous confirmed-OHLCV warmup → realtime normalized-trade lifecycle
• preserve positions, equity, pending orders, Pine/TA state, request.security() state, and partial timeframe aggregation across the handoff
• evaluate resting broker orders on every normalized trade while retaining the default close-only strategy calculation cadence
• expose a provider-neutral four-field trade POD through the stable C ABI:
  • source event timestamp
  • optional monotonic normalized sequence
  • executed price
  • quantity in the configured symbol's volume units
• keep provider authentication, transport, symbol mapping, and provider-only metadata outside the engine
• skip quiet intervals outside the configured syminfo session instead of manufacturing tradable bars across market closures
• expose the lifecycle through the Python ctypes harness and contiguous mmap replay tooling
• add a dedicated cdocs streaming guide and runnable tutorial

Why

TradingView strategies load confirmed historical bars and then continue on realtime data without replacing the strategy instance. PineForge previously only exposed one-shot backtests, so switching data sources could not preserve the complete broker and script state.

This PR introduces the engine-owned lifecycle needed by a separate data layer such as pineforge-data. Any community provider adapter can normalize its payload into pf_trade_tick_t; the engine remains deterministic, dependency-free, and unaware of HTTP, WebSockets, credentials, or vendor schemas.

It also establishes the realtime execution foundation needed for later alert(), order-fill alert, and JSON webhook work. Alert delivery itself is intentionally not part of this PR.

Validation

Current head (1226c64):

• ctest --test-dir build --output-on-failure: 85/85 passed, including provider-neutral ABI coverage, session-aware gap handling, and test_tutorial_stream
• ./scripts/run_corpus.sh: 252/252 executed; 239 excellent, 12 strong, 1 existing anomaly; no moderate/weak/minimal results
• python3 scripts/check_c_abi_runtime.py: passed
• bash docs/build.sh: generated without warnings
• Python syntax checks for the ctypes harness, both stream corpus runners, and tutorial: passed
• git diff --check: passed

Existing 24×7 raw-trade replay evidence for the streaming lifecycle:

• 756/756 sessions completed
• 756/756 input/script bar-count parity
• 750/756 trade-count parity
• 729/756 exact ordered trade structure
• 99.866844% weighted ordered trade-structure parity across 4,463,189 compared trades

The long replay used three session starts (2025-03-29, 2025-03-30, and 2025-03-31), each ending exclusively at 2025-05-01. Each strategy/session received one contiguous strategy_stream_push_ticks(pointer, count) call.