feat(tools): add universal tool result budgeting for large outputs

[chinesepowered]

Apply universal tool result budgeting so all tools benefit from large output truncation and disk persistence, not just shell and MCP.

TLDR

Adds universal tool result truncation in the core tool scheduler. Previously, only run_shell_command and MCP tools truncated oversized outputs. Now ALL tool results (grep, glob, read_file, edit, etc.) are subject to the same budgeting: when output exceeds the configured threshold (default 25,000 chars / 1,000 lines), the full result is saved to a temp file and a truncated head/tail preview with a file path pointer is sent to the model.

Screenshots / Video Demo

N/A — no user-facing UI change. The behavior is the same truncation users already see for shell output, now applied universally.

Dive Deeper

• Single integration point in coreToolScheduler.ts (line ~1481) — after execute() returns but before convertToFunctionResponse() builds the chat history entry
• Uses the existing truncateToolOutput() utility from utils/truncation.ts which handles threshold checking, head/tail extraction, disk persistence, and telemetry logging
• Tools that already self-truncate (shell, MCP) pass through harmlessly since their output is already under the threshold — no double-truncation
• Respects existing user-configurable thresholds: tools.truncateToolOutputThreshold and tools.truncateToolOutputLines
• Only applies to string llmContent — media results (images, PDFs) are unaffected

Modified files:

• packages/core/src/core/coreToolScheduler.ts — Added universal truncation + import

Reviewer Test Plan

1. Run a grep that returns thousands of matches — verify output is truncated with head/tail and a temp file path
2. Read a very large file — verify same truncation behavior
3. Run a shell command with large output — verify it still works (no double truncation)
4. Check the temp file referenced in truncation message actually contains full output
5. Use read_file on the temp file path — verify full content is accessible
6. Run existing scheduler tests: npx vitest run src/core/coreToolScheduler.test.ts (all 53 pass)

Testing Matrix

	🍏	🪟	🐧
npm run	❓	✅	❓
npx	❓	❓	❓
Docker	❓	❓	❓
Podman	❓	-	-
Seatbelt	❓	-	-

Linked issues / bugs

New feature to prevent context window waste from oversized tool outputs.
Fixes #2818