restore performance of integer shift operators.

These were calling convert, which generally works, but is slow as
it checks the conversion from Int to Int32, somewhat unnecessarily.

base/operators.jl

@@ -111,9 +111,9 @@ const .≠ = .!=
 <<(x,y::Int32) = no_op_err("<<", typeof(x))
 >>(x,y::Int32) = no_op_err(">>", typeof(x))
 >>>(x,y::Int32) = no_op_err(">>>", typeof(x))
-<<(x,y::Integer) = x << convert(Int32,y)
->>(x,y::Integer) = x >> convert(Int32,y)
->>>(x,y::Integer) = x >>> convert(Int32,y)
+<<(x,y::Integer) = x << (y % Int32)
+>>(x,y::Integer) = x >> (y % Int32)
+>>>(x,y::Integer) = x >>> (y % Int32)
 
 # fallback div, fld, and cld implementations
 # NOTE: C89 fmod() and x87 FPREM implicitly provide truncating float division,