feat(constants)!: switch URLs to v0.9.0 layout + add MODEL_REGISTRY

[msluszniak]

Description

Switches every URL constant in the library to the restructured HF layout
under resolve/v0.9.0 and adds the typed MODEL_REGISTRY accessor.

URL refresh

All URLs now follow <model>_<size>_<backend>_<precision>.pte, files
sit under per-size and per-backend directories on HF:

• modelUrls.ts — 170 URL refs rewritten. The lfm2_5_350m_xnnpack_8w4da.pte typo is corrected to _8da4w.pte.
• ocr/models.ts — CRAFT detector URL + CRNN per-language URL switch to the new <lang>/xnnpack/crnn_<lang>_xnnpack_fp32.pte shape.
• tts/models.ts — Kokoro consts re-rooted to <size>/xnnpack/kokoro_<size>_<component>_xnnpack_fp32.pte.
• tts/voices.ts — voices/ and phonemizer/ asset paths kept in place; only the tag value bumps.
• versions.ts — VERSION_TAG → resolve/v0.9.0. NEXT_VERSION_TAG removed. PREVIOUS_VERSION_TAG = resolve/v0.8.0 retained for the @deprecated Llama QLoRA aliases.

MODEL_REGISTRY

New constants/modelRegistry.ts exports a typed accessor grouped by
capability (LLM / VLM / CLASSIFICATION / OBJECT_DETECTION /
SEMANTIC_SEGMENTATION / INSTANCE_SEGMENTATION / STYLE_TRANSFER /
SPEECH_TO_TEXT / TEXT_EMBEDDING / IMAGE_EMBEDDING / IMAGE_GENERATION /
VAD). Each entry is callable with { quant, backend }:

MODEL_REGISTRY.LLM.LLAMA3_2_3B                   // default (quantized)
MODEL_REGISTRY.LLM.LLAMA3_2_3B({ quant: false })  // base

Object access returns the default config; calling resolves the
requested variant. backend is accepted in the type signature for
forward-compat — the library still picks via Platform.OS at module
load. Per-backend selection lands in a follow-up.

Deprecations

• LLAMA3_2_3B_QLORA, LLAMA3_2_1B_QLORA — @deprecated; the .pte
  files stay at v0.8.0, and the constants still resolve those URLs.
  Use LLAMA3_2_*_SPINQUANT going forward.

Introduces a breaking change?

• Yes
• No

URL paths under ${VERSION_TAG} change — any code that hardcoded
resolve/v0.8.0 URLs through the constants keeps working only if it
read the constants at runtime (the constant values themselves are
updated). Historical tags continue to resolve old paths, so apps pinned
to a previous library version are unaffected.

The flat MODEL_REGISTRY = { ALL_MODELS: {...} } export in
modelUrls.ts is removed; the new MODEL_REGISTRY from
constants/modelRegistry.ts is the replacement. The internal
URL→name lookup (getModelNameForUrl) is preserved.

Type of change

• New feature (change which adds functionality)
• Other (chores, tests, code style improvements etc.)

Tested on

• iOS
• Android

yarn typecheck clean across the monorepo. Runtime behaviour validated
against the migrated HF state (every URL resolves at v0.9.0).

Testing instructions

yarn typecheck
# Or in the package:
cd packages/react-native-executorch && yarn typecheck

In application code:

import { MODEL_REGISTRY } from 'react-native-executorch';
console.log(MODEL_REGISTRY.LLM.LLAMA3_2_3B.modelSource);
console.log(MODEL_REGISTRY.LLM.LLAMA3_2_3B({ quant: true }).modelSource);

Related issues

#431
#612

Checklist

• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have updated the documentation accordingly
• My changes generate no new warnings

[barhanc]

This code is duplicated in several files. It should probably be factored out to some common utilities file. However, in general I would be in favour of not using the model accessor as both a value and a function.

[barhanc]

I don't think we should make the accessor behave like both a value and a function and just make the user explicitly call e.g. MODEL_REGISTRY.LLM.LLAMA3_2_3B() for default config (perhaps with some stylistic changes like changing the names to lowercase to indicate these are getters and not constants; the use pattern in the user code would be something like const LLAMA3_2_3B = models.llm.llama3_2_3b()). I feel the current approach might generate some problems e.g. when comparing models as in example apps using sameValue workaround.