Use ordinal when checking for hex prefix

[ellismg]

In general lingustic comparsion is slower than ordinal comparision.
When using ICU for globalization, the lingustic StartsWith code path
actually ends up doing a lot of work inside ICU itself. While we'll try
to make things better (see dotnet/corefx#3672) for common cases like
ASCII only, if code doesn't need ligustic comparision the best practice
is to not request it.

BaseNumberConverter was using a lingustic StartsWith to detect a hex
prefix when trying to parse a number but it is better served by just
always using ordinal comparisons.

This results in a ~35% QPS improvement in some very barebones ASP.NET
scenarios (like the one outlined in richardkiene/CoreCLRWebAPISample) on
my local machine.

[ellismg]

@ianhays or @krwq Can you please take a look?

/cc @richardkiene

[ianhays]

LGTM. Nice little fix :) Thanks, Matt.

[viktor-evdokimov]

That might be naive question, but shouldn't

 else if (this.AllowHex && text.StartsWith("0x", StringComparison.OrdinalIgnoreCase)
                                      || text.StartsWith("&h", StringComparison.OrdinalIgnoreCase))

be

else if (this.AllowHex && (text.StartsWith("0x", StringComparison.OrdinalIgnoreCase)
                               || text.StartsWith("&h", StringComparison.OrdinalIgnoreCase)))

i.e. check for prefix if flag is true?

[JesperTreetop]

@ianhays I'm wondering what @ojosdegris wondered too. This is intact in the current version.

[ianhays]

Well it's consistent with what was there before this commit, but it does seem like an issue.

[JesperTreetop]

The .NET Framework version does this too so this may be bug compatibility.

[JesperTreetop]

Then again, if bug compatibility is to be maintained, 0X should be matched without the hex flag.