feat(fuse): add max_readahead_kb config; fix clippy::field_reassign_with_default

[Copilot]

The CI "build" job was failing at the clippy step because newly introduced tests used post-construction field assignment on a Default::default() instance, triggering clippy::field_reassign_with_default (denied as warning).

Root cause

// Before — clippy error
let mut conf = FuseConf::default();
conf.max_readahead_kb = Some(0);  // error: field assignment outside of initializer

// After — correct form
let mut conf = FuseConf { max_readahead_kb: Some(0), ..Default::default() };

Changes

curvine-common/src/conf/fuse_conf.rs

• Add max_readahead_kb: Option<u32> field with doc comment (sysfs path corrected to read_ahead_kb)
• Reject Some(0) in init() — value must be positive when set
• Fix test initialization to use struct literal syntax

curvine-fuse/src/session/bdi.rs (new)

• Linux-only helper that resolves the FUSE mount's BDI device via /proc/self/mountinfo and writes kb to /sys/class/bdi/<maj>:<min>/read_ahead_kb
• Retries up to BDI_RETRY_COUNT × BDI_RETRY_DELAY_MS ms for the sysfs entry to appear post-mount; all failures are best-effort (log + continue)
• No-op stub for non-Linux targets

curvine-fuse/src/fs/curvine_file_system.rs

• During FUSE init, raise max_readahead to max(negotiated, max_readahead_kb * 1024) so the kernel cap doesn't silently shrink reads

curvine-fuse/src/session/fuse_session.rs

• After mount setup, call bdi::apply_max_readahead_kb for each mount point when the option is set

etc/curvine-cluster.toml

• Document the new # max_readahead_kb = 1024 option under [fuse]

[lzjqsdd]

@copilot fix merge conflicts