feat: harden base MATLAB workflows and app governance

[Pluze]

What changed

• Replaced hard Image Processing Toolbox dependencies with MATLAB-compatible labkit.image APIs and added base-MATLAB dependency/shadow-workflow guardrails.
• Corrected CIC and VT batch recomputation, clarified delay units, centralized scientific constants, and exposed FLIR calibration provenance.
• Added shared draggable rectangle and anchor interactions, including DIC rigid point matching and app ROI governance.
• Made MATLAB CI progress, active-test state, shard logs, timeouts, and assumption-filtered results traceable and reliable.
• Rebuilt CHANGELOG.md as structured schema-v1 evolution records, reconstructed every tracked launcher/facade/app version chain, and exposed selected-app history in the launcher.

Why

Base MATLAB users need the same supported workflows as toolbox-rich developer machines. Scientific calculations and conversions also need explicit provenance, reusable interaction contracts, and diagnostics that identify the real test or shard responsible for failures instead of producing opaque CI exits.

User and developer impact

• Image, DIC, electrochemistry, thermal, and ROI workflows no longer require optional toolboxes for their supported baseline paths.
• CIC batch output uses each file's own detected timing and displays delay with an explicit time unit.
• Launcher users can inspect complete structured version history for the selected app.
• Assumption skips remain visible in test summaries but no longer fail a shard.

Validation

• buildtool baseMatlab: passed 5/5.
• buildtool headless: passed 267 tests; the product-ownership test remained a visible assumption skip where appropriate.
• buildtool gui: passed 65/65 in hidden GUI mode.
• Final strict buildtool changed with LABKIT_ENFORCE_VERSION_BUMPS=1: passed DIC non-GUI 64/64 and GUI 2/2.
• Focused changelog guardrail: passed 5/5; launcher suite: passed 14/14 before the catalog/history split, followed by catalog/history 3/3.

Not verified manually

Interactive visual feel, real hardware acquisition, and real laboratory files were not exercised. Automated GUI checks used hidden synthetic workflows.