Add integer to Half and BFloat16 conversion rounding tests

[tannergooding]

Adds test coverage proving that integer->Half and integer->BFloat16 conversions round correctly and do not double-round through float. This is part of #112474.

HalfTests previously had no integer-conversion tests at all, and BFloat16Tests only covered Int32/UInt32. The integer conversions are already correct in the current tree (BFloat16 uses direct RoundFromSigned/RoundFromUnsigned, and Half saturates below 2^17 so no integer can observably double-round); these tests guard that behavior against regressions.

What's covered

• Half: Int32/UInt32/Int64/UInt64, exercising the 11-bit round-to-even boundary (2048/2049/2050/2051, 4097/4098/4100), the finite/overflow boundary (65504/65519/65520 -> inf), and large 64-bit magnitudes -> inf.
• BFloat16: Int64/UInt64/Int128/UInt128, exercising the mid-range double-rounding boundary (0x101_0000/0x101_0001/0x100_FFFF - the values that are inexact in float, so a two-step conversion would round the wrong way) plus high-magnitude round-to-even boundaries near 2^62, 2^63, 2^100, and 2^127, and MaxValue/MinValue.

Validation

Every expected bit pattern was cross-checked against the single-step correctly-rounded Half.Parse/BFloat16.Parse oracle (integer decimal strings are exact). The discriminating case was confirmed to differ from a two-step result: (BFloat16)0x101_0001 is 0x4B81 (correct, direct) whereas (BFloat16)(float)0x101_0001 is 0x4B80 (double-rounded) - so these tests fail if the conversion is ever routed through float.

Test-only change. Both classes pass (1605 Half, 1571 BFloat16, 0 failures).

Note

This PR description was generated by GitHub Copilot.

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-numerics
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

This PR expands System.Runtime numeric tests to cover explicit integer-to-Half and integer-to-BFloat16 conversions, with expected results specified by exact bit patterns to validate correct rounding behavior (including cases that would fail if an implementation ever double-rounded via float).

Changes:

• Add Half explicit conversion test vectors for Int32/UInt32/Int64/UInt64, including round-to-even boundaries and finite/overflow boundaries.
• Add BFloat16 explicit conversion test vectors for Int64/UInt64/Int128/UInt128, including mid-range double-rounding discriminators and high-magnitude round-to-even boundaries.

Show a summary per file

File	Description
src/libraries/System.Runtime/tests/System.Runtime.Tests/System/Numerics/BFloat16Tests.cs	Adds explicit integer conversion rounding tests (including double-rounding discriminator cases) for 64-bit and 128-bit integers.
src/libraries/System.Runtime/tests/System.Runtime.Tests/System/HalfTests.cs	Adds explicit integer conversion rounding/overflow tests for 32-bit and 64-bit signed/unsigned integers.

Copilot's findings

• Files reviewed: 2/2 changed files
• Comments generated: 0