[peerbit-org]

On FUSE 3, hostAsgnCconfig zeroes the struct fuse_config that libfuse passes to the init callback:

static inline void hostAsgnCconfig(struct fuse_config *conf, bool direct_io, bool use_ino)
{
	memset(conf, 0, sizeof *conf);   // <- wipes the parsed configuration
	conf->direct_io = direct_io;
	conf->use_ino = use_ino;
}

In libfuse 3 the high-level init receives &f->conf, the live configuration that fuse_new already filled from defaults and -o options. Clearing it sets entry_timeout, attr_timeout and negative_timeout to 0 for the whole mount, so -o entry_timeout=…,attr_timeout=… are silently ignored on FUSE 3. The kernel then issues a LOOKUP for every path component on every path syscall, and a GETATTR for every stat.

This PR removes the memset and keeps setting only the fields cgofuse owns (direct_io, use_ino). FUSE 2 builds are unaffected, because the function is empty there.

Measured on Linux (Ubuntu 24.04, libfuse 3.14, kernel 6.17) with a cgofuse-based filesystem, counting filesystem Getattr callbacks:

	stock	with this change and -o entry_timeout=0.1,attr_timeout=0,negative_timeout=0
stat of a 2-component path	3 callbacks	1
git status on 2,000 files	14,084	4,974

• Passing the same -o options to the stock build leaves the counts unchanged, which shows they were being wiped.
• A libfuse 2 build of the same filesystem (defaults 1/1/0) matches the patched FUSE 3 build at 1/1/0.

One behaviour change for FUSE 3 users: without explicit -o options they now get libfuse's defaults (entry and attr timeouts of 1 s) instead of 0. That matches FUSE 2 and macFUSE, and filesystems that need uncached attributes can pass -o attr_timeout=0.