Bound decompression dictionaries with static size limits

docs/API.md

@@ -327,6 +327,39 @@ await writer.WriteAllAsync("C:\\source", "*", SearchOption.AllDirectories, cance
 
 ## Common Options
 
+### Decompression dictionary limits
+
+`SharpCompress.Common.DictionarySizeLimits` contains process-wide static size limits
+in bytes. Set them once at application startup, not around individual requests:
+
+```csharp
+// Opt in only when trusted archives require larger dictionaries.
+DictionarySizeLimits.Rar5 = 256L << 20;
+```
+
+| Static property | Default maximum |
+| --- | --- |
+| `Rar4` | 4 MiB (the RAR4 format maximum) |
+| `Rar5` | 64 MiB |
+| `Lzma` (raw LZMA, LZip, ZIP, 7z) | 64 MiB |
+| `Lzma2` (XZ, 7z) | 64 MiB |
+| `Ppmd` (ZIP, 7z, RAR models) | 64 MiB |
+| `ZStandard` (streaming windows) | 128 MiB |
+
+Limits are read directly by the shared codecs before dictionary/model allocation,
+so Archive, Reader, direct decoder, sync, and async paths use the same settings.
+No reader options or decoder overloads are needed. Settings require positive values;
+invalid assignments throw `ArgumentOutOfRangeException` without changing the setting.
+Oversized dictionaries throw `InvalidFormatException`; Zstandard reports
+`ZstdException` with `ZSTD_error_frameParameter_windowTooLarge`. Zstandard's minimum
+configurable window limit is 1 KiB. Format and engine restrictions still apply.
+
+These are dictionary/window or model limits, not total-memory, output-size, or
+concurrency limits. Pool rounding and allocator overhead can increase actual memory.
+Shared PPMd allocators also apply the model limit during compression. Fixed/bounded
+codecs retain their format-defined bounds. Custom providers must enforce their own
+limits. Zstandard's one-shot `Unwrap` APIs retain their existing output-size bounds.
+
 ### ReaderOptions
 
 Use preset properties and fluent helpers for common configurations: