fix: improve non-interactive TTY detection in install.sh

[mango766]

Fixes #3294

Checks

• I am happy for maintainers to push small adjustments to this PR, to speed up the review cycle
• I have checked that there are no existing pull requests for the same thing

Problem

The install script's non-interactive detection uses { true </dev/tty; } to check if a terminal is available. This only tests whether /dev/tty can be opened, not whether it is actually a functional terminal. In containers, CI, and headless environments (like the Ona setup described in #3294), /dev/tty may exist as a device node that the shell can open, but isn't backed by a real terminal. This causes:

1. The detection to incorrectly conclude the session is interactive
2. The subsequent read </dev/tty calls to fail with cannot open /dev/tty: No such device or address
3. The final atuin setup </dev/tty call to hang indefinitely

Fix

Two changes, both in install.sh:

1. Stronger TTY detection — Replace { true </dev/tty; } with ( exec </dev/tty && [ -t 0 ] ). This opens /dev/tty as stdin inside a subshell, then checks it's actually a terminal with test -t. The subshell ensures the redirect doesn't affect the parent process.

2. Suppress stderr on read fallbacks — Add 2>/dev/null to the two read </dev/tty calls. They already have || var="n" fallbacks, but without stderr suppression, failed redirects produce noisy error messages on shells like dash.

[greptile-apps]

Greptile Summary

Fixes false-positive interactive detection in install.sh for environments where /dev/tty exists but isn't backed by a real terminal (containers, CI, headless setups). The fix is correct and POSIX-compatible.

• Replaces { true </dev/tty; } with ( exec </dev/tty && [ -t 0 ] ) to verify /dev/tty is actually a terminal, not just openable
• Adds 2>/dev/null to read </dev/tty calls to suppress noisy errors on dash and similar shells

Important Files Changed

Filename	Overview
install.sh	Strengthens TTY detection and suppresses stderr on read fallbacks; changes are correct and POSIX-compatible.

_{Last reviewed commit: "fix: improve non-int..."}

[mentalisttraceur]

Thank so much for taking care of this!

[mentalisttraceur]

I recommend not suppressing errors from the two reads.

Think about it this way: those error messages were the reason why we were able to diagnose the issue in my bug so quickly.

Without them, it would just be an unexplained hang.

My issue is fixed with a better TTY check, but if we hit some other problem during read, the error messages would be helpful.