Terminal capability detection reworked into a probe-and-settle model

What

• Synchronized-output detection was restructured into reason-carrying helpers; checks for TMUX, CLAUDE_CODE_FORCE_SYNC_OUTPUT, and JetBrains terminals now live together, returning a reason string (like TERM_PROGRAM=...) instead of a plain boolean. The old Konsole-version allowlist entry and the old runtime terminal self-query for sync-output support were removed.
• A new Capabilities tracker treats extendedKeys, synchronizedOutput, and kittyKeyboard as either 'settled' (known) or 'pending' (a guess awaiting a probe response), recording a source/reason for each. Probing now also issues a CSI ?u query for kitty-keyboard/extended-key support, alongside the existing DECRPM 2026 sync-output probe, and treats DECRPM status 3 (not just 1 or 2) as 'supported'.
• The function that used to shell out to git rev-parse --show-object-format to detect a repo's sha256/sha1 object format was rewritten to instead decide whether synchronized terminal output is active (checking CLAUDE_BG_BACKEND, CLAUDE_CODE_FORCE_SYNC_OUTPUT, and the new capability state). The git object-format check itself moved into its own dedicated helper.

Why Detecting terminal features by actually probing and reading the terminal's reply, rather than guessing from TERM or hardcoded version allowlists, makes Claude Code's terminal rendering more reliable across a wider range of terminal emulators.