Generate AI embedding vectors for AI expression topics

[bobular]

This PR integrates embedding API calls to generate semantic vector representations for each topic's headline and summary text. The embeddings are computed by concatenating the headline and summary with double newlines, then calling the embedding service to produce 512-dimensional vectors. A new embedding_vector field (number array) is optionally added to the topic response payload for use in client-side similarity calculations and visualization. The "Other" topic does not get a vector.

The token usage handling was overhauled while we were at it (to avoid applying the much higher chat model costs). The $0.02/M input tokens cost has been hardcoded in DailyCostMonitor

TO DO:

• Add config mechanism for enabling/disabling the embedding functionality - now via a makeTopicEmbeddings prop in the JSON request payload to the SingleGeneAiExpressionReporter - defaults to false of course.
• Remove requirement for OPENAI_API_KEY to be set if makeTopicEmbeddings is false
• Add hard-coded configurable support for extended reasoning in ClaudeSummarizer