Unify agent prompt handling and update chat input UX

[wgqqqqq]

Summary

• remove the GPT-only agentic prompt template so all models use the default agentic prompt
• thread current working directory through tool contexts and tighten file path resolution around the active workspace
• cap Read tool output with lower default line limits plus a total character budget, and surface truncation explicitly
• include the local chat input UX and locale updates that were already in the working tree

Verification

• cargo test -p tool-runtime truncates_when_total_char_budget_is_hit
• cargo test -p tool-runtime reads_multiple_lines_when_budget_allows
• cargo test -p bitfun-core always_uses_default_prompt_template
• cargo check -p bitfun-desktop

Notes

• This PR intentionally includes all local changes that were present in the worktree, including the web UI input updates.