[Filestore] Proper byte size calculation in ReadNodeRefs upon ListNodes

[debnatkh]

The ReadNodeRefs function in

nbs/cloud/filestore/libs/storage/tablet/tablet_database.cpp

Line 619 in cf42914

bytes += refs.back().Name.size();

has a FIXME regarding byte size calculation. The current implementation uses an incorrect size estimation.

There are two problems with that:

1. When we precharge nodeRefs, we take into consideration the entire row size, but when we iterate over them, we only take into consideration the name size, meaning that we definitely will not precharge enough bytes and force the PrepareTx restart:

   nbs/cloud/filestore/libs/storage/tablet/tablet_actor_listnodes.cpp

   Line 135 in cf42914

   if (!PrechargeNodeRefs(

2. Different callers may need different size calculations depending on their target, specifically:

• DirEntry: FUSE fuse_dirent size (FUSE_DIRENT_ALIGN(24 + namelen))
• DirEntryPlus: FUSE fuse_direntplus size (includes fuse_entry_out ~120+ bytes + dirent)
• (optional) Legacy: name.size() only (original behavior)
• (optional) RefSize: Full TNodeRef struct size (sizeof(ui64) * 4 + Name.size() + ShardId.size() + ShardNodeName.size())

This is needed, because API explicitly specifies: https://libfuse.github.io/doxygen/structfuse__lowlevel__ops.html#af621ef96ffc2a71af8c7768c109b880e

Send a buffer filled using fuse_add_direntry(), with size not exceeding the requested size. Send an empty buffer on end of stream.