Task: Make the tenant roster a workspace

The tenant roster reads as a page rather than a box in the middle of one. The table is no longer wrapped in a card that repeats the title and the count, the title, the count and the primary action sit in a page header, and the shell gives a list-shaped screen more room than it gives a form.

• The roster draws the table as the page, with no card around it.
• The page header carries the title, the count and a primary action that opens New tenant, at any row count.
• The empty roster states the state once and does not repeat the header's action beneath it.
• The shell bounds a screen at one of two named widths, and the screen chooses: a rail, form or dialog takes the column, a list or grid takes the workspace.
• A screen that has not chosen takes the column, so a form across a wide display is not the mistake that happens by omission.
• A journey's banner is still bounded, and the roster table is not.
• Every new behaviour is proved by a test, and the layout was read in a browser at a wide display.

1. Two widths, named for their shape. One bound for every screen was wrong in both directions, which is how the banner and the table came to disagree. The widths live in one module and are named for the shape they suit.
2. The screen chooses, and the default is the safer mistake. The shell owns the bound so a journey cannot forget it, and the default is the column: a form running the width of a wide display is the harder defect to read.
3. Then the page. The card comes off, the header gains the primary action, and the empty state stops repeating it.

Every number this task claims about a tree, with the command that produced it and the commit it was measured at. A number without both is not a measurement, because a clean-up record describes a tree and trunk moves under a branch: a consumer count, a gate's component count and a suite's size all change with no code changing, and the reader of a bare number cannot tell which tree it came from. Where the raw output matters, put it in an evidence/ file beside the task and name that file in the Command column.

Manual QA scenarios (scaffolded via compass add test_scenario) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

The roster is a workspace, and the shell has two bounds instead of one.

• The table is the page. The card around it carried the title and the count the page header already states, and made the table narrower than the screen. It is gone. The header carries the title, the count and a primary action that opens New tenant at any row count, and the empty roster states the state once instead of repeating the header's action beneath it.
• Two widths, named for their shape. column for a rail, a form or a dialog, and workspace for a list or a grid, both in shell/layout.ts. The shell owns the bound so a journey cannot forget it, and the default is the column, because a form across a wide display is the harder defect to read.
• The banner is still bounded and the table is not. On a 1600px display the journey's banner draws 794px wide, as before; on a 1920px display the roster table draws 1600px, where the card had held it to 1100px with a dead margin on either side.

Proved live: the harness signs in, reads the roster, and checks that the table is not inside a card, that the page offers the journey that creates a tenant, and that the table takes the width it was given. Screenshot in evidence/tenant_roster.png.

What this task does not do, and what it hands to the review:

• The workspace bound is a guess. 1600px with five columns is airy. The number is one constant, and the review is where it gets settled.
• No search, no filters, no row actions. The subject has no filter, and the journey a row action would open is not built. Both stay out rather than appear as controls that do nothing.
• The sidebar. Deliberately not touched. It is the next decision, and the result of this task is what it should be decided against.