util-genai: ToolInvocation._apply_finish never calls the completion hook, so tool arguments/results bypass redaction hooks

Version: opentelemetry-util-genai 1.1b0 (_tool_invocation.py:120).

What happens. InferenceInvocation._apply_finish and WorkflowInvocation._apply_finish
call _call_completion_hook(...) before ending the span, so a CompletionHook registered through
the opentelemetry_genai_completion_hook entry point (the documented way to add redaction or
upload) sees inference and workflow content. ToolInvocation._apply_finish writes
gen_ai.tool.call.arguments and gen_ai.tool.call.result and ends the span without calling the
hook, so tool content bypasses every hook, including redaction.

Reproduce. Register a hook that records every call; run handler.tool("t") with
arguments={"token": "sk-..."}; the hook is never called and the attribute reaches the exporter.

Suggested fix. Call _call_completion_hook from ToolInvocation._apply_finish with the
arguments and result as inputs/outputs (or add a dedicated hook method for tools), consistently
with the other invocation types.

Downstream workaround (forgehawk SDK). forgehawk's own tool emitters (api.py, the LangChain subclass) redact
arguments and results directly before assignment (redaction.redact_tool_value); third-party
tool spans stay unredacted until this is fixed.

Found while building the forgehawk Python SDK (ZioSec) on top of opentelemetry-util-genai; happy to open a PR for the suggested fix if maintainers agree with the direction.

Start at _tool_invocation.py:120 and compare ToolInvocation._apply_finish with the corresponding finish methods on InferenceInvocation and WorkflowInvocation. Verify how _call_completion_hook receives inputs and outputs, then confirm that tool arguments and results pass through the hook before the span ends. Done means registered hooks observe both values, including for redaction.