feat(debugger): add statement stepping and polish script editor

[jrepp]

Summary

Adds a usable, session-scoped HypeTalk debugger with statement breakpoints, Step Into/Step Over semantics, pollable debug-port automation, modal-error inspection/dismissal, and a script editor whose text and controls remain readable, correctly sized, and synchronized with external script changes.

Context

PR #292 establishes the in-process tracing/debugger implementation used by the live Hype debug bridge and the follow-on MCP surface in PR #286. The earlier handler-entry tracing worked, but authoring and external acceptance testing needed executable-line breakpoints, real nested-call stepping, execution scoping, nonblocking automation, and a reliable script-editor UI.

No .hype document shape, persistence, keychain, provider, or stack file-access behavior changes. Debugger and debug-operation state remains process/session-only. The local debug bridge remains a privileged local Unix-socket boundary and all wrapped mutation calls continue through their existing mutation gates.

Changes Made

• Added structured HypeTalk handler tracing, runtime profiling, scoped variables, watchpoints, pause/resume controls, and live debugger state.
• Captured executable statement source locations and validated script-editor breakpoints against handler declarations and executable lines; blank/comment/terminator lines are rejected.
• Scoped stepping and breakpoint-hit bookkeeping by dispatch and handler execution identity so concurrent runtimes cannot consume each other's debugger state.
• Implemented Step Into across nested handler calls and Step Over at the next statement in the current/calling handler.
• Excluded nested debugger pause time from caller runtime-budget accounting so external stepping does not create false frame-pressure reports.
• Added a bounded process-local debug-operation registry and debug/startOperation, debug/pollOperation, and debug/forgetOperation RPCs. Terminal results expire after five minutes; the registry is capped at 128 operations.
• Refactored hype_wait_for_debugger_pause and hype_step_script_execution_and_wait to return operation UUIDs immediately instead of holding callers open.
• Routed stdio MCP button clicks and message dispatch through pollable operations so scripts halted at breakpoints cannot block the MCP caller.
• Added hype_list_alerts and hype_dismiss_alert, including a modal-run-loop-safe socket path that can inspect and clear document-open errors while normal main-actor work is suspended.
• Scoped script-editor/debugger window and menu automation to the originating stack, including multi-document editor identity and source navigation.
• Made open script editors adopt external hype_set_script changes and corrected displayed debug-document activation.
• Reworked debugger/editor split-pane sizing, fixed command-palette header contrast, and preserved readable light/dark AppKit text, selection, gutter, and syntax colors.
• Updated architecture/debug-bridge documentation and added focused debugger, operation-lifecycle, automation, appearance, hosted-layout, and timing regressions.

Testing & Verification

• Ran relevant automated tests on the final source:
  • scripts/test.sh focused debugger/MCP operation filters — 24 tests in 3 suites passed.
  • swift test --no-parallel --filter HypeCoreTests --filter HypeCLITests — 3,361 tests in 349 suites passed, including interpreter fuzz/property coverage.
  • node --check Tools/hype-mcp-server/bin/hype-mcp.js passed.
• Performed live Unix-socket acceptance testing:
  • A debugger wait returned an operation UUID immediately.
  • debug/pollOperation remained responsive while AppKit displayed an application-modal invalid-document error.
  • The modal was enumerated and dismissed through the debug port, then the operation reached a structured terminal result.
  • Generic start/poll/forget completed successfully for debug/getState.
• Performed live stdio MCP acceptance testing: attach, hype_start_debug_operation, and hype_poll_debug_operation completed successfully against the deployed app.
• Visually confirmed the debugger/editor layout, external script refresh, correct multi-document scoping, and readable command-palette/editor appearance.
• Built/deployed/launched the app: ./script/build_and_run.sh --deploy passed; /Applications/Hype.app was relaunched and verified.
• git diff --check passed.

Screenshots / GIFs

Checklist

• I read the relevant parts of architecture.md and decisions.md.
• I considered persistence, network, keychain, file-system, script-execution, and AI-tool safety risks.
• I added or updated tests, or explained why tests are not needed.
• I updated documentation for behavior, setup, architecture, persistence, or workflow changes.
• I did not stage or rewrite .hype stack documents unless required by this change.
• I noted any follow-up work or known limitations.

Known limitations: debugger breakpoints, watchpoints, pauses, trace entries, and pollable operation results are intentionally process/session state and are not persisted into user stacks. Pending operations remain until their underlying work completes; completed and failed results expire after five minutes.