fix: raise nofile soft limit to the hard limit at startup

[V0UP3R]

Problem

On hybrid-GPU systems where the panel renders through the NVIDIA driver, cosmic-panel crashes every few hours with EMFILE, or freezes invisible, because it exhausts the default soft limit of 1024 open file descriptors (#643).

The journal signature at the moment of death is:

cosmic-session[3332]: error marshalling arguments for import_timeline: dup failed: Too many open files
cosmic-session[3332]: Error trying to flush the wayland display: Too many open files (os error 24)
cosmic-panel[987403]: com.system76.CosmicAppletAudio: exited with code 137
[... every embedded applet killed, panel exits with code 1, restart counter grows]

Root cause

The NVIDIA explicit-sync path leaks sync_file fds inside the driver blob (NVIDIA/egl-wayland#196). On my system (AMD 680M iGPU + RTX 3050, driver 580.173.02) the panel accumulates 15-20 anon_inode:sync_file fds per minute of active use; strace -k shows the leaked fds are created by dup3 calls inside libnvidia-eglcore.so.580.173.02 and never closed. That leak is not fixable in cosmic-panel, but the default 1024 fd ceiling turns it into a crash within hours.

cosmic-comp already raises its nofile soft limit to the hard limit at startup for exactly this class of problem (src/utils/rlimit.rs), but cosmic-panel is spawned by cosmic-session, not cosmic-comp, so it never gets that benefit and sits at the plain session default.

Fix

Raise the soft limit to the hard limit (1048576 by default on Pop!_OS) during panel startup, mirroring cosmic-comp. This turns "crashes every few hours" into a ceiling that takes weeks of uptime to reach. Applet processes inherit the raised limit, which helps them too since they render through the same driver stack (cosmic-app-library#371 shows the same sync_file pile-up). The panel does not exec arbitrary user commands, so the raised limit stays within COSMIC components.

Testing

On the affected machine (Pop!_OS 24.04, COSMIC, AMD 680M + RTX 3050 hybrid, NVIDIA 580.173.02), A/B test with an artificially low soft limit to make the ceiling reachable quickly:

• Unpatched panel started with ulimit -Sn 300: dies in under 20 seconds with dup failed: Too many open files, exit code 1 — the exact production failure signature.
• Patched panel started with the same ulimit -Sn 300: raises itself to 1048576 (verified via /proc/<pid>/limits), climbs past the 300 fd mark without any error, and runs until the test timeout ends it.

cargo build and cargo clippy pass; the only warnings are pre-existing.

Note: this is a mitigation for the driver-side leak, not a fix for it — the leak itself is tracked upstream at NVIDIA/egl-wayland#196.

• I have disclosed use of any AI generated code in my commit messages.
• I understand these changes in full and will be able to respond to review comments.
• My change is accurately described in the commit message.
• My contribution is tested and working as described.
• I have read the Developer Certificate of Origin and certify my contribution under its conditions.

Assisted by an AI coding tool; I have reviewed and fully understand the change and am able to maintain it and respond to review.