Add TOKEN_ECONOMY_GUIDANCE.md for budgeting AI spend and model choice

[awrobel-gd]

• Introduced a new document detailing token economics and a repeatable budgeting model for AI spend.

• Added a section on predicting monthly costs based on usage patterns and model choices.

• Included a worked example to illustrate the budgeting process for engineers using AI in their development workflow.

• more recommendation is needed — keep models in the same technology level (product generation), comparable LLM size and capabilities

  • Rewrote the comparability recommendation: the variant fleet must be one product generation, within ~2× price band, similar coding capability (with a plain rule-of-thumb), and explained why — a mixed fleet makes divergence unreadable (spec vs. capability gap) and per-variant cost erratic.

• Which models to use as LOW / MEDIUM / HIGH

  • Added a "Which models to use for LOW / MEDIUM / HIGH" subsection: per-tier job, what to pick, why, and the current default, plus swap-in guidelines (don't mix generations within a tier, HIGH ≥ every MEDIUM model, fewer MEDIUM = cheaper/weaker signal, provider format).