[Filestore] Optimization epic

[qkrorlqr]

Important:

• detecting garbage sooner to keep index size smaller which speeds-up all index operations, helps to store less garbage and speeds-up tablet startup: [Filestore] Compaction/Cleanup/FlushBytes optimization #95
• optimizing read throughput for multiple clients: [Filestore] Make Filestore read throughput scalable with respect to the number of clients #324
• optimizing write throughput for multiple clients: [Filestore] Make Filestore write throughput scalable with respect to the number of clients #539
• optimizing localdb compaction policy which will speed up all index operations: [Filestore] Change compaction policy inside IndexTablet #688
• optimizing sequential reads: [Filestore] Support read-ahead for sequential reads #852
• optimizing full tablet load time to start processing all write/update operations sooner for large filesystems: [Filestore] Optimize CompactionMap loading time #1128
• increasing background ops' throughput to keep index size smaller and speed-up all index operations: [Filestore] Batch Cleanup & Compaction #1129
• optimizing all inode index-related operations: [Filestore] Node index cache #1146
• optimizing DescribeData/ReadData (mostly): [Filestore] MixedBlocks LRU Cache #1226
• load balancing: [Filestore] Proper tablet balancing #1175
• single FS scalability by increasing the number of tablets per FS: [Filestore] Multitablet filesystems #1350
• optimizing IOPS for small write requests: [Filestore] Store fresh blocks (and maybe fresh bytes?) outside of local db - directly in a channel #1518
• optimizing write throughput for unaligned requests: [Filestore] Support ThreeStageWrite for unaligned writes #1673
• optimizing inefficient writing patterns like sequential write using tiny requests: [Filestore] writeback-cache for files which are open for writing by only one client #1751
• optimizing open/close latency: [Filestore] non-blocking open/close #1541
• optimizing Compaction once again: [Filestore] Avoid Compaction loops #1922

Nice to have:

• increasing read iops by scaling DescribeData request processing across multiple cores: [Filestore] Offload TDescribeDataRequest processing via TReadAheadCache to separate actors #1131
• optimizing unaligned writes: [Filestore] Don't write the aligned part of a large unaligned WriteData request to FreshBlocks #1288 - covered by [Filestore] Support ThreeStageWrite for unaligned writes #1673 in the ThreeStageWrite datapath
• providing the client with the correct blocksize [Filestore] Provide client with correct blocksize #1289
• [Filestore] listdir should not do excessive getnodeattr when possible #1585