[endrix]

Phase 5 of the declutter plan, second step: two places where the platform still carried a product's choices as its own defaults. Cut from main; independent of the other open Dialogram PRs (only CHANGELOG.md can conflict).

Breaking for a profile that relied on the defaults. The next release bumps DIALOGRAM_API_VERSION to 0.9 (the check requires an exact major.minor before 1.0), so a product built for 0.8 fails loudly instead of silently losing its palette and navigation. Not bumped here: four open PRs change the API surface, and the 0.8 bump was made in its release commit.

Ordering: ship after endrix/wfpy-ide#23 and streamblocks/streamblocks-ide#9, which declare what this PR stops providing. Against today's main, with those two PRs, the palette and navigation are unchanged: a one-off test built each product's old and new input through createSidecarDiagramProfile and WorkflowToolPaletteItemProvider and compared the whole palette, identical down to item ids, icons and args.

Palette

tool-palette-provider.ts built Entities from two hard-coded lists: wfpy's Task / Tool / Agent / Viewer / Workflow by default, calpy's Actor / Network behind useAlternateEntityPalette, with platformEntityItems: false to switch both off. Entities are now the profile's entityPaletteItems, in its order, and nothing else. useAlternateEntityPalette and platformEntityItems are @deprecated, accepted and ignored (in StorageRuntimeOptions, DiagramStorageOptions and SidecarProfileInput), so sv, vhdl and ndf, which set platformEntityItems: false, compile and behave as before. The toolkit stops forwarding them. A profile with no rows shows no Entities header (before, platformEntityItems: false with no rows left an empty one). chisel-ide sets supportsElementCreation: false (no palette); mlir-viewer builds no palette.

Navigation

SidecarProfileInput.navigation defaulted to 'python': every product without the line got the Python navigation provider and the .py caller scan. The default is 'none'; createPythonNavigationProvider() and the two Python workspace walks in the model source's analysis seam run only for navigation: 'python'. The rest of the analysis seam (overlay matching, default unit, definition ranges) is untouched.

Source MIME type

The source the agent receives defaulted to text/x-python; it is text/plain unless the profile names one. sv, vhdl and ndf name theirs; the two product PRs name text/x-python.

Tests

Server: the palette test now pins that Entities are exactly the declared rows with their ids, and that the retired flags yield an empty Entities category. Toolkit: the profile tests name navigation: 'python' where they expect the provider and check the default builds none; the analysis test checks the default answers empty and 'python' scans.

Typecheck 5/5, neutrality 5/5, server 441, toolkit 326 (the known local wizard timeouts reproduce on main), extension build green.

From the review

Taken: the Python MIME default (the toolkit's defaults now agree with navigation 'none'); no empty Entities header; the port-description edit reverted, since a palette item carries no description and it never reached the screen; the stale navigation docs; identity checks for the wired scan functions; one provider per test case; the breaking-change framing and the 0.9 bump at release. Not here: the create-node handler still reads wfpy's agentTask/viewerTask args, which goes with the agent stack (phase 5e).

https://claude.ai/code/session_01EToj6RJRMSSHQiqGNMU9wv