daemon: resolve the shell once instead of probing /etc/shells on every exec

[mu-hashmi]

Is your feature request related to a problem? Please describe.

Every toolbox exec, code-run, session create, and PTY spawn resolves the user's shell by spawning sh -c "grep '^[^#]' /etc/shells" before launching the actual command. That doubles the execve count on the daemon's hottest paths for a value that cannot change while the sandbox runs. Under degraded filesystem conditions the probe is also the first thing to fail, silently downgrading users to sh even when bash/zsh are installed.

Describe the solution you'd like

Read and parse /etc/shells in-process (no subprocess) and cache the resolved shell for the daemon's lifetime. A failed resolution should not be cached, so a transient failure does not pin the sh fallback forever.

Acceptance criteria:

• Shell resolution performs no subprocess spawn
• Successful resolution is computed once and reused across exec/session/PTY paths
• Failed resolution falls back for that call only and retries on the next call
• Existing shell preference order is unchanged (zsh > bash > $SHELL > sh)

Describe alternatives you've considered

Caching the current subprocess result (keeps an unnecessary execve on first use and the grep dependency). Per-request resolution with a TTL (complexity without benefit; /etc/shells is static in a running sandbox).

Additional context

Small change, but it removes one syscall-heavy step from every interactive operation and makes the exec path more robust exactly when the filesystem is unhappy.