v0.1.10

Extension contract versions are unchanged: world 5, provider 5, bot 6.

Core

• A batch that ends over hardTokens right after the previous batch was force-handed-off at the same limit no
  longer hands off again. Core logs one error (hardTokens, estTokens, model) and stops count-based round-end
  finishes and forced handoffs at that limit until a check sees the count back under it or the limit changes.
  Before, an endpoint whose limit came out as 0 handed off after every batch and the handoff note woke the next
  batch, so the bot looped without end. Handoffs on an upstream "input too long" error and manual handoffs are
  unchanged (#201, Pal-AI-Lab/Coopanion#100).

Providers and console

• An endpoint spec that sets both contextWindow and maxTokens must have contextWindow > maxTokens. Saving
  such an endpoint or making it current is refused, and the endpoint editor marks the field, also when an already
  saved endpoint is opened (#201).
• A listed model without a context window takes the window the provider instance reports
  (ProviderInstance.contextWindow), so the editor fills it in instead of asking for it by hand. llamacpp
  listings are kept as listed, so fetching the list does not probe each model (#201).
• The diagnostics bundle has an endpoint section: the endpoint name, its provider module and the spec's model,
  thinking, reasoningEffort, temperature, maxTokens and contextWindow; no address, key or options
  (#201).