Fix deserialization of primitive arrays on Big Endian systems

[medhatiwari]

Description

This PR fixes deserialization issues in PrimitivesArrayReader.cs which caused 51 test failures in the MongoDB.Bson.Tests suite on Big Endian systems (e.g., IBM s390x).

Problem

The deserialization logic relied on BitConverter.ToXxx(...) which assumes Little Endian memory layout. On Big Endian architectures, this results in incorrect values for arrays of primitive types such as Int64, Single, Double, and Decimal, leading to test failures.

Fix

• Replaced unsafe BitConverter usages with BinaryPrimitives.ReadXxxLittleEndian(...) for endian-safe reads.

• Covered conversions for:

    -- float, double (via BitConverter.Int64BitsToDouble(...))
  
    -- decimal (via Decimal128.FromIEEEBits(...))
  
    -- char (casted from ushort)
  
    -- All numeric types (e.g. int, long, ushort, etc.)
  

Results:

51 test failures resolved.

Example fixed tests:

MemorySerializerTests.ReadonlyMemory_should_roundtrip_special_values<Int64>

MemorySerializerTests.Memory_should_roundtrip_special_values<Single>

BinaryVectorSerializerTests.BinaryVectors_should_roundtrip<float>

Pre-requisites:

• s390x system (e.g., IBM Z, RHEL on s390x)

• .NET 8.0 SDK installed

cc: @giritrivedi

[medhatiwari]

hi @rstam , Can you please review this?

[rstam]

We will discuss this at our next triage meeting. Not sure whether we intend to support Big Endian systems or not.