Fix Llama model fallback logic in modelCapabilities.ts

[hyperfinitism]

This pull request updates the fallback logic for Llama models in modelCapabilities.ts.

Llama 3

The current fallback logic for Llama3 behaves as follows (in order):

1. if modelName includes llama3 → use llama3
2. else if modelName includes llama3.1 → use llama3.1
3. else if modelName includes llama3.2 → use llama3.2
4. else if modelName includes llama3.3 → use llama3.3

With this order, any modelName that includes llama3.1, llama3.2 or llama3.3 matches clause (1), and is captured by the generic llama3 before it can fall back to the more specific llama3.1, llama3.2, or llama3.3 variants as intended.

This PR fixes the issue by moving clause (1) to the end, so that the specific llama3.1 / llama3.2 / llama3.3 take precedence over the generic llama3 match.

Llama 4

Currently a modelName matching llama4-maverick incorrectly falls back to llama4-scout. This PR also fixes this issue.