fix: handle oversized ollama embedding inputs

[Helweg]

Summary

Fix oversized built-in Ollama embedding inputs by aligning the observed token budget with runtime behavior and pooling provider-sized subchunks back into a single stored embedding per semantic chunk.

Changes

• align the built-in nomic-embed-text budget and docs with the observed ~2048-token effective limit
• add provider-level Ollama truncation retry coverage as a last-resort safety valve for oversize requests
• split oversized embedding texts into multi-part requests in the indexer, pool the part vectors before storage, and persist recovered failed-batch retries back to SQLite

Testing

How were these changes tested?

• Unit tests added/updated
• Manual testing performed
• Build passes (npm run build)
• Typecheck passes (npm run typecheck)
• Tests pass (npm run test:run)
• Lint passes (npm run lint)

Release Labels

• Added at least one release category label (feature, bug, performance, documentation, dependencies, refactor, test, chore, or skip-changelog)
• Added at most one semver label (semver:major, semver:minor, semver:patch) when needed

Related Issues

Fixes #58