Refactor Zig code to use array type instead of bare arrays all over the place

[emil-e]

Safer code is better code. Chose not to use Zig builtin types because the ergonomics are poor and require the buffer to be separate.

[emil-e]

@dakra Any input or should I just merge?

[dakra]

For me the code looks great. Some claude review comments, do with them what you like:

• ghostel--has-wide-chars is now set per-row instead of once per render. Old code aggregated has_wide_chars across rows and called env.call2(intern("set"), …, t) exactly once after the loop (gated on dirty != DIRTY_FALSE). New code calls it inside insertAndStyle for every dirty row that contains a wide cell. Functionally equivalent (idempotent set), but it adds an intern + call2 per such row. Minor, but trivially recoverable: keep the old return-the-flag pattern, or stash the

fixed_array_list.zig

• Consider std.BoundedArray(T, N) from the stdlib — it's the same idea (inline buffer + len, append/slice/constSlice/resize, error.Overflow). The PR description says builtin types have "poor ergonomics," but BoundedArray is exactly the shape you built. If there's a concrete reason it doesn't fit (missing method, Zig 0.15.2 API gap), worth a one-line comment in the file; otherwise dropping the helper would save ~180 LOC.
• unusedCapacitySlice simplifies to self.buffer[self.len..capacity]. The self.len + self.unusedCapacity() form is correct
  but indirect.
• Typo: // Returns a slice of the reamining capacity. → remaining (src/fixed_array_list.zig:39).
• items() and constItems() have identical doc comments; one says "mutable" the other "read-only" would help.
• Missing blank line between the last two tests (constItems and resize: shrinks…), src/fixed_array_list.zig:171.