5.2.0

• Fixed decoding of some valid four-byte data pointers.
• Fixed truncated or out-of-range results when decoding oversized integers.
• Added NativeAOT and trimming support for C# model deserialization. The NuGet
  package now includes a source generator for constructor-based and
  property-based models, including models with annotated properties inherited
  from abstract base records.
• Added reflection-free source-generated activation for supported generic
  collection and dictionary interfaces and concrete types, including closed
  collection types used directly with Find<T> and FindAll<T>.
• Reused immutable source-generated activation metadata across readers, reducing
  the time and allocation cost of the first model lookup on a new reader.
• Improved lookup performance and reduced allocations when creating collections
  through reflection.
• Enabled trim, AOT, and single-file compatibility analysis.
• Added the MMDBSG001 through MMDBSG016 diagnostics, which report model
  shapes the generator cannot support so that they are caught at build time
  rather than after trimming or an AOT publish. They are reported by default in
  any project that declares models; set the MaxMindDbAotDiagnostics MSBuild
  property to false to turn them off.
• Added the MaxMind.Db.SourceGeneratorSupport and MaxMind.Db.GeneratedMember
  public types. They exist for generated code to call and are not intended to be
  used directly.
• Fixed a [MapKey(..., true)] member of a non-nullable value type throwing
  during property-based activation instead of keeping its default. This affected
  the reflection path before this release and is now consistent across both.
• Added decoding limits to prevent crafted databases from consuming excessive
  time and memory. Each lookup and metadata read allows at most 65,536 decoded
  values, 512 nesting levels (including pointer follows), and 2 MiB of combined
  string, bytes, uint32, uint64, and uint128 payload. Exceeding a limit or
  encountering a pointer cycle throws InvalidDatabaseException. These limits
  reject some previously accepted databases, including those with a string or
  bytes value larger than 2 MiB. Available stack space can impose a lower
  nesting limit.
• Pointers that point directly to other pointers now throw
  InvalidDatabaseException.
• Truncated or out-of-bounds data reads now throw InvalidDatabaseException
  instead of ArgumentOutOfRangeException. This also prevents incorrect decoded
  values on netstandard2.0. Update error handlers to catch
  InvalidDatabaseException.