Migrate LiteLsm reads to iterator API, remove class-based SkipList

- Remove SkipList.cs (class-based) and SkipListType enum; only StructSkipList remains
- Make GetIterator public with ReadAll() yielding IEnumerable<(TKey, TValue)>
- Fix LiteLsmIterator: buffer span bug, batch continuation via excludeStart
- Fix ReadBatchInto to handle segment-to-memtable boundary transitions
- Add excludeStart support through MemCpyRawRange and SegmentManager (O(log N) + O(1))
- Fix GetKeyIdxInDataLayer to return successor index for not-found keys
- Fix SegmentManager FileShare for concurrent read/write access
- Update tests and benchmarks to use iterator API
- Add WaitForPendingFlush to write benchmark for fair comparison
- Add concurrent stress test (50K inserts with parallel point + range readers)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: 2 people

benchmarks/StreamDB.Benchmarks/ReadBenchmarks.cs

@@ -290,8 +290,10 @@ public int LiteLsm_RangeRead()
         // LiteLsm has no secondary indexes — query ~1000 entries to match other single-index reads
         long startPi = BasePi + 10_000;
         long endPi = startPi + 1000;
-        var results = _liteLsm.QueryRange(startPi, endPi).ToList();
-        return results.Count;
+        using var iter = _liteLsm.GetIterator(startPi, endPi);
+        int count = 0;
+        foreach (var _ in iter.ReadAll()) count++;
+        return count;
     }
 
     [Benchmark]
@@ -300,8 +302,10 @@ public int LiteLsm_MultiKeyRead()
         // Comparable to multi-index reads: query ~4000 entries
         long startPi = BasePi + 10_000;
         long endPi = startPi + 4000;
-        var results = _liteLsm.QueryRange(startPi, endPi).ToList();
-        return results.Count;
+        using var iter = _liteLsm.GetIterator(startPi, endPi);
+        int count = 0;
+        foreach (var _ in iter.ReadAll()) count++;
+        return count;
     }
 
     [Benchmark]
@@ -310,16 +314,20 @@ public int LiteLsm_InMemory_RangeRead()
         // All data in memtable — no disk I/O
         long startPi = BasePi + 10_000;
         long endPi = startPi + 1000;
-        var results = _liteLsmInMemory.QueryRange(startPi, endPi).ToList();
-        return results.Count;
+        using var iter = _liteLsmInMemory.GetIterator(startPi, endPi);
+        int count = 0;
+        foreach (var _ in iter.ReadAll()) count++;
+        return count;
     }
 
     [Benchmark]
     public int LiteLsm_InMemory_MultiKeyRead()
     {
         long startPi = BasePi + 10_000;
         long endPi = startPi + 4000;
-        var results = _liteLsmInMemory.QueryRange(startPi, endPi).ToList();
-        return results.Count;
+        using var iter = _liteLsmInMemory.GetIterator(startPi, endPi);
+        int count = 0;
+        foreach (var _ in iter.ReadAll()) count++;
+        return count;
     }
 }

benchmarks/StreamDB.Benchmarks/WriteBenchmarks.cs

@@ -183,6 +183,7 @@ public void LiteLsm_SequentialWrites()
         {
             _liteLsm.Put(1_000_000 + i, payload);
         }
+        _liteLsm.WaitForPendingFlush();
     }
 
     private static void TryDelete(string? path)

src/StreamDB/BorrowedMemory.cs

@@ -0,0 +1,7 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace StreamDB
+{
+}

src/StreamDB/LiteLsm/IMonotonicSkipList.cs

@@ -0,0 +1,15 @@
+namespace StreamDB.LiteLsm;
+
+public interface IMonotonicSkipList<TKey, TValue> where TKey : unmanaged, IComparable<TKey> where TValue : unmanaged
+{
+    int Count { get; }
+    //bool TryGetMinKey(out TKey key);
+    //bool TryGetMaxKey(out TKey key);
+    bool ProbablyContainsKey(TKey key);
+    void Clear();
+    IEnumerable<KeyValuePair<TKey, TValue>> GetAll();
+    ReadOnlySpan<byte> GetRawDataLayer();
+    void InsertMonotonic(TKey key, TValue value);
+    bool TryGetValue(TKey key, out TValue value);
+    (bool, int) MemCpyRawRange(TKey fromKey, TKey endKey, Span<(TKey, TValue)> buffer, bool excludeStart = false);
+}