Add some tests and missing float operations to NDArrat

[Quafadas]

No description provided.

[Quafadas]

@copilot The unaryOps general path all do double to float and back conversions.

Why? Is this necessary? If it’s necessary, can we fold it inline into the general ops helper to aid performance?

[Copilot]

@copilot The unaryOps general path all do double to float and back conversions.

Why? Is this necessary? If it’s necessary, can we fold it inline into the general ops helper to aid performance?

The explicit .toDouble calls were redundant — Java/Scala widens Float to Double implicitly when calling transcendental functions like Math.exp, Math.log, etc. (since the JVM only provides double-precision versions of those). The .toFloat at the end is still necessary to store the result back into a Float array.

For Math.abs, Java has a float overload (Math.abs(float) → float), so no conversion is needed at all.

Fixed in fb45462:

• Made unaryOpGeneral inline with inline f: Float => Float (matching the existing flatUnaryOp pattern), so the function body gets inlined at each call site instead of going through a virtual dispatch
• Removed all redundant .toDouble calls from the general-path lambdas