💡 flex-shrink and flex-basis

Start with the failing test in test/flex-shrink.test.ts, then read clay/clay.h around lines 2318 and 2358-2401 to understand sizing and overflow distribution. Trace the wire path through src/clayterm.c:365 and ops.ts:16 and 216. Done means weighted shrink and basis behavior works end to end while the default preserves no-shrink behavior and the repro renders all three glyphs.

Is your feature request related to a problem?

When children overflow their parent along the main axis, we can't absorb the overflow by shrinking sized children. The solver only redistributes free space across grow children and excludes fixed/percent from compression, so an overflowing row keeps every child at its declared size and pushes trailing children off-grid. A width-10 row with two fixed(6) children plus a trailing fit() lays A at width 6, B at 6, and C at x=12—off the 10-wide grid—so C never renders, giving "A B " instead of all three glyphs.

Describe the solution you'd like

A child should support a flex-basis as the solver's starting main size plus a shrink weight controlling how much overflow it absorbs. shrinkWeight (exact name TBD) should mirror the CSS flex-shrink property and default to 0, preserving today's no-shrink behavior; basis mirrors CSS flex-basis and seeds the main size before the solver runs. For the scene above, marking both width-6 children shrinkable by 1 distributes the 3-column overflow (6→5, 6→4), landing C at x=9 and rendering "A B C".

open("a", {
  layout: {
    width: {
      ...fixed(6),
      basis: 6,
      shrinkWeight: 1,
    },
    height: fixed(1),
  },
});

Describe alternatives you've considered

A JS-side workaround means measuring intrinsic sizes, computing the negative-free-space split, and re-emitting each child as a fixed() of the shrunk width before render. That duplicates the solver in the bindings, can't see wrapped-text or nested intrinsic sizes the way Clay does, and breaks the moment a parent resizes.

Additional context

Failing test case on nm/repro/flex-shrink (test · diff). Today A and B stay at width 6 and C lands at x=12. Likely spans the Clay solver and the wire format: clay/clay.h:2318 skips FIXED/PERCENT from resizableContainerBuffer, clay/clay.h:2358-2401 runs the sizeToDistribute < 0 loop with no per-child shrink weight, src/clayterm.c:365 decode_axis has no basis/shrink word, and ops.ts:16 packAxis plus the ops.ts:216 SizingAxis union would need a basis/shrink-aware shape.