#1688 setup: let the swift family take IQ3_S (#1651)

Descrição

Fixes #1651.

`MODELS["IQ3_S"]["families"] = ("qwen",)` dates from when Swift 1.5 had no IQ3_S; UkisAI published one on 2026-10-08. Dropping the key restores the `("qwen", "swift")` default, so `--family swift --model IQ3_S` is offered and accepted.

## Verified against the Hub before writing this

`ukisai/Swift-1.5-Qwen3.8-Flash-Next-GSQ-RCO-GGUF` serves `Swift-Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-0000{1,2}-of-00002.gguf` — the family `file` pattern verbatim; sizes 44,900,816,736 + 38,842,920,384 bytes and both SHA-256s match the issue exactly. Shard count (2) and the shared BF16 `mmproj` are unchanged, so the download/pack path is the one the other Swift tiers already use.

## Test plan

- `tools.test_setup_choices.SwiftIq3s`: the size appears in the swift choices and the shard names match the Hub listing.
- `tools.test_setup_choices` + `golden` + `risk`: 86 tests, all pass.

Not tested: an actual download/run of the tier (83.7 GB) — same trust level as the other shared sizes; setup verifies the files by name and size after the download either way.