fix: Log warning when input text is truncated to context limit

[echobt]

Description

This PR addresses the issue of silent token truncation when input text exceeds the context size limit. Previously, tokens were dropped without any indication to the user, potentially leading to incomplete embeddings and misleading search results.

Changes

• Added a warning log to src/core/embeddings.rs that prints to stderr when truncation occurs.
• The warning includes the original token count and the truncated (context limit) count.

Verification

• Verified by code inspection that the truncation logic now includes a warning side-effect.
• Existing tests pass.