[LLM] Add tokenizer-native token accounting and cost projection

[Simonsbs]

Goal

Strengthen token-efficiency measurement so it predicts real LLM cost/latency decisions.

Scope

• Add provider-tokenizer-based counts (where available) alongside whitespace proxy.
• Report prompt, completion, and total tokens per task and per pass.
• Add cost projection mode (optional) using configurable token price table.

Acceptance

• Reports include both proxy and tokenizer-native counts.
• Token efficiency trends are charted in wiki.
• Methodology doc updated with caveats.