[FR] Add granular control to exclude pages/blocks from embeddings (like .gitignore for AI indexing)

### Description

AppFlowy currently automatically **sends all workspace content to the configured embeddings provider** to enable semantic search and AI-powered features.

However, there is no mechanism to exclude specific pages, documents, or blocks from being included in the embedding/indexing process.

This limits control over what data is processed by AI systems, especially for users integrating AppFlowy with external or self-hosted AI pipelines (e.g. local LLM servers, vector databases, or custom memory systems).

### 🛠️ Suggested implementation approach

This feature could be implemented in a lightweight and extensible way without changing the core embedding pipeline:

### 1. Metadata-based exclusion (recommended baseline)
Allow pages or blocks to include a simple flag:

`ai_index: false`

or

`embedding: disabled`

This would allow the embedding pipeline to skip content during chunking.

### 2. Tag-based exclusion rules
Support tags such as:
- no-ai
- private
- no-index

which would automatically exclude tagged content from embedding generation.

### 3. Optional .aiignore file (workspace-level rules
Similar to .gitignore, allowing patterns such as:
`/private/*
/journal/drafts/*
*.secret`

This would enable flexible workspace-wide exclusion rules.

4. Hierarchical policy resolution
In case of conflicts, rules could follow a priority system:

`Block-level rule > Page-level rule > Workspace-level rule`

### Impact

The absence of granular embedding control introduces architectural and usability constraints for advanced and privacy-sensitive use cases:

- Global indexing without exclusion control: **all workspace content is embedded by default,** with no ability to opt out specific pages or blocks
- Lack of data boundary separation: **no distinction** between AI-searchable content and private or non-indexed information
- Privacy and governance limitations: users cannot enforce selective embedding policies for sensitive or internal data
- Reduced compatibility with external AI architectures: limits integration with custom memory systems, vector databases, and agent-based workflows
- Scalability constraints for advanced setups: makes it difficult to adopt AppFlowy in production-grade or hybrid AI environments where data control is required

### 💡 Why this matters

For modern AI-driven workflows, embedding pipelines are often treated as a controlled data layer rather than a global automatic process.

🚨 Without granular control, AppFlowy behaves as a fully inclusive indexing system, which reduces flexibility for advanced users and self-hosted deployments.

### Additional Context

_No response_