fix: null dangling downstream input pointer on node removal

[striderZA]

Summary

Fix a use-after-free segfault: deleting an upstream node (e.g. an RF ADC feeding a PFB Channelizer) freed the node's SignalNode synchronously mid-draw_ui(), leaving downstream components' node().inputs[] pointers dangling until the next frame's rewire — any widget dereferencing that pointer during the same frame's draw (PFBChannelizerWidget) crashed.

Related issue

Closes #37

Type of change

• Bug fix
• New feature
• Refactor
• Documentation
• Build / CI

Test plan

• cmake -B build -G Ninja && cmake --build build
• ctest --test-dir build --output-on-failure
• Manual verification steps:
  • New regression test (tests/test_issue37_pfb_input_removal.cpp) builds Generator→ADC→PFB, wires inputs via update_dsp(), then invokes the same onRemoveNode callback the UI uses on the ADC node, and asserts the PFB's node().inputs[0] is nulled immediately rather than left dangling.
  • Confirmed the test fails (dangling non-null pointer) against the pre-fix code (git stash on app.cpp/app.h), and passes with the fix.
  • Full existing suite green: tests.exe (217 cases / 65522 assertions), test_extensions, test_component_authoring, test_attenuator, test_combiner.
  • Full tiny-rf-simulator app target builds and links cleanly.

Checklist

• My code follows the existing code style (see .clang-format)
• I ran clang-format -i on changed files (via scripts/format.sh)
• I added or updated tests where appropriate
• Existing tests still pass

Details

RfSimulatorApp::update_dsp()'s signal-routing loop is extracted into RfSimulatorApp::rewireInputs(). onRemoveNode now calls rewireInputs() immediately after ComponentRegistry::remove(), so every surviving component's node().inputs[] reflects the current graph topology (nulled for any severed source) before the rest of that frame's draw_ui() continues rendering. This closes the dangling-pointer window for every widget that reads node().inputs[] directly during draw (PFB channelizer grid, and similarly the inspector panel's input-frequency readout), not just the reported PFB crash.

Also documents the new safety contract in app/AGENTS.md.