optimize: lock-free Usenet FUSE pipeline

[javi11]

Five targeted changes to reduce lock contention on the hot read path
and eliminate the prefetch scheduling stall:

• segmentRange.current → atomic.Int64: GetCurrentIndex() is now
  lock-free; called on every downloadManager iteration, this removes
  an RLock/RUnlock from the tightest inner loop.

• UsenetReader.totalBytesRead → atomic.Int64: eliminates a
  sync.Mutex lock/unlock pair inside the innermost Read() byte-copy
  loop, the single highest-frequency lock acquisition in the codebase.

• UsenetReader.nextToDownload → atomic.Int64: removes mutex from
  GetBufferedOffset() and simplifies downloadManager state tracking.

• Replace 50ms polling timer with segmentConsumed channel: when
  Read() advances past a segment via rg.Next(), it sends a non-blocking
  signal on a buffered channel. downloadManager wakes immediately
  rather than waiting up to 50ms, reducing stall latency at segment
  boundaries (especially visible at high bitrates / fast connections).
  A 1s fallback timer guards against missed signals.

• sync.Pool for download buffers: reuses *bytes.Buffer across
  segment downloads to reduce GC allocation pressure when 30+
  segments are in-flight concurrently.

• FUSE MaxReadAhead default: 128KB → 4MB. Usenet segments are
  ~750KB decoded; a 4MB kernel readahead window allows the kernel to
  pipeline multiple full-segment reads without stalling on round-trips,
  keeping the prefetch queue saturated.