feat: API key edit, protected mode, system terminal and files

[nfebe]

Summary

• API keys can now be edited and carry per-deployment access levels (read/write/admin), capped by the owning user's level. The owner-resolution path no longer returns a 401 when an admin API key creates a new key (which was logging users out in the UI).
• New protected-mode controls for deployments and for the global system terminal: configurable blocked actions, blocked command rules, and an enable/disable switch.
• New system-wide endpoints for the file manager and terminal, gated by new permissions, with chmod and file creation support and a cheap files-info variant for fast initial render.

Closes #117
Closes #122
Closes #123

[sourceant]

Code Review Summary

This PR introduces significant security and administrative features, including editable API keys with granular deployment access, a 'Protected Mode' for deployments to prevent accidental destructive actions, and system-wide file and terminal management. The core architectural change is moving from simple list-based deployment access to a level-based map.

🚀 Key Improvements

• Added granular access levels (read/write/admin) per deployment for API keys.
• Implemented a 'Protected Mode' framework to block specific actions (delete, update env) or specific commands via regex/string matching.
• Introduced system-level file management and terminal access gated by admin permissions.

💡 Minor Suggestions

• Use more efficient database queries for finding admin users instead of filtering the whole list in memory.
• Improve terminal command reconstruction to handle terminal control characters better.

🚨 Critical Issues

• Potential for command injection in system terminal if shell meta-characters bypass the blocklist.
• Performance risk with recursive disk usage calculations on the system root.

[sourceant]

Review complete. See the overview comment for a summary.

[sourceant]

The system terminal executes commands using sh -lc. If the input raw is not properly sanitized, an attacker might be able to inject commands despite the protectedCommandBlocked check (e.g., using command substitution or redirection).

Suggested change

	}
	command := parts[0]
	args := parts[1:]
	if command == "ll" {
	command = "ls"
	args = append([]string{"-la"}, args...)
	raw = strings.Join(append([]string{command}, args...), " ")
	}
	if command == "la" {
	command = "ls"
	args = append([]string{"-A"}, args...)
	raw = strings.Join(append([]string{command}, args...), " ")
	}
	if command == "cd" {
	target := session.cwd
	if len(args) > 0 {
	target = resolveSystemTerminalPath(session.cwd, args[0])
	}
	info, err := os.Stat(target)
	if err != nil {
	return "", err
	}
	if !info.IsDir() {
	return "", fmt.Errorf("not a directory: %s", target)
	}
	session.cwd = target
	return "", nil
	}
	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
	defer cancel()
	cmd := exec.CommandContext(ctx, "sh", "-lc", raw)
	// Consider using a more restricted shell or ensuring 'raw' does not contain malicious shell meta-characters
	cmd := exec.CommandContext(ctx, "sh", "-lc", "--", raw)