Fix OpenVINO inference engine integration issues from PR #38

[Copilot]

PR #38 added OpenVINO inference support but had critical bugs preventing functionality and several safety issues.

Critical Fixes

Config path bug: _init_config() read from cfg.get("engine_cfg", {}) instead of cfg directly, causing all YAML settings to be ignored:

# Before: settings never applied
engine_cfg = cfg.get("engine_cfg", {})  # Wrong nesting
_set("PERFORMANCE_HINT", engine_cfg.get("performance_hint"))

# After: reads from correct path
_set("PERFORMANCE_HINT", cfg.get("performance_hint"))

Missing import: os.cpu_count() used without importing os, causing immediate NameError.

None injection: _set() with lambda cast could add None values to OpenVINO config dict:

# Before: could set config["INFERENCE_NUM_THREADS"] = None
config[k] = cast(v)  

# After: guards against None
casted_value = cast(v)
if casted_value is not None:
    config[k] = casted_value

Safety Improvements

• Guard os.cpu_count() returning None before min() comparison
• have_key() now checks "framework" level consistently with get_character_list()
• Specific exception handling (AttributeError, TypeError, KeyError) instead of bare except
• Defensive attribute access with hasattr/getattr before calling .value.splitlines()
• Safe .get() pattern throughout to prevent KeyError

Style Fixes

• YAML indentation: 2-space → 4-space to match onnxruntime block
• Consistent single-line _set() calls except where lambda spans multiple lines
• EOF newline in __init__.py

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.