Currently, the env var list collapses or splits rows based on whether the value is identical across environments:

• Same value in dev + prod -> one row
• Different values -> two rows

This breaks the core mental model. The number of rows in the list no longer equals the number of variables. Scanning becomes unreliable because you can't tell at a glance how many variables you have or whether a variable exists in a given environment.

Proposed behavior:

• Always show one row per variable name
• Clicking a row opens a modal
• The modal shows each environment with its current (masked) value and an input to update it
• All edits to one variable happen in one place

This keeps the list scannable and predictable. The "collapse identical values" optimization can still happen inside the modal if space is a concern, but it shouldn't affect the list itself.