Inaccurate brand name detection in prompt targeting, leading to misclassification of unbranded prompts.

### NiubiGEO version or commit

3aa01de (current main)

### What happened?

The prompt targeting check uses a plain substring search for the target name. This can mark an unbranded question as branded when the name is short or also appears inside another word.

For example, with target name `Go` and domain `go.dev`, the question `What is a good deployment tool?` is marked with `targetIncluded: true` because `go` is the first two letters of `good`. That prompt is then treated as branded in the audit categories and can distort the natural discovery and brand awareness metrics.

The same check is used when prompts are generated and when they are edited in the UI.

### Steps to reproduce

1. Start an audit with target name `Go` and domain `go.dev`.
2. Add or generate the prompt `What is a good deployment tool?`.
3. Inspect the prompt's `targetIncluded` value or the category shown in the plan.
4. It is classified as containing the target brand even though the standalone brand name does not appear.

The current helper returns:

```ts
promptContainsTarget("What is a good deployment tool?", target) // true
```

### Expected behavior

A target name should match as a standalone term when it is a Latin word, while still allowing normal substring matching for CJK names where word boundaries do not apply. The example above should be treated as an unbranded prompt.

### Environment

OS: Windows 11
Node: 22+
Commit: 3aa01de