IBX-11401: Fixed embedding default model inheritance and added field name fallback prefixes

[mikadamczyk]

🎫 Issue	IBX-11401

Related PRs:

• https://github.com/ibexa/connector-gemini/pull/2
• https://github.com/ibexa/taxonomy/pull/399
• IBX-11401: Added generic dense vector dynamic fields for dimension-based embedding suffixes solr#111
• https://github.com/ibexa/elasticsearch/pull/73

Description:

This PR contains two related core changes needed for Gemini embedding integration and backward-compatible search field naming.

Fixed default_embedding_model inheritance across SiteAccess scopes

Embeddings parser no longer defines a semantic default value for default_embedding_model. A scalar default defined in the semantic config node was being materialized into every SiteAccess/group scope during mapSetting(). This prevented proper inheritance from system.default overrides (e.g., Gemini config set in project config), because scopes such as site / admin_group ended up with explicit text-embedding-ada-002.

• Removed semantic defaultValue('text-embedding-ada-002') from Embeddings parser
• Kept the default in core default settings (ibexa.site_access.config.default.default_embedding_model), so BC is preserved

Added generic field name fallback prefixes in FieldNameGenerator

FieldNameGenerator now supports configurable fallback prefixes used when no explicit field type mapping exists. New embedding models should not require adding a dedicated common.yml mapping entry for every model suffix. We need a generic way to turn ibexa_dense_vector_<suffix> into backend field suffix

• Existing explicit mappings (e.g., OpenAI suffixes) remain unchanged and still take precedence
• The inheritance fix restores the expected system.default override behavior without changing existing explicit SiteAccess/group overrides

For QA:

Documentation:

[barw4]

Aside from Konrad's remarks

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud