Precision loss When Reading in CSV data

### Summary
CSV fields not auto- reading In values correctly (losing precision)

### Scenario
I want to automatically read some Integer Id's into my dataframe. These integers are quite large and are losing precision.

### Cause
https://github.com/dotnet/machinelearning/blob/main/src/Microsoft.Data.Analysis/DataFrame.IO.cs
***Method: GuessType()***
We don't make any attempt to read Int, Long or Double Types here. Hence the code-flow is truncating the long integers.
I guess since this is primarily a ML framework, we favour floats, and therefore want to avoid the performance overhead of more complex type inference?

### Suggestions
**Numeric Preference**
It may be nice if the dev can specify a preferred numerical type to the Read_Csv method. i.e double, long, float, int. I suppose the disadvantage maybe unnecessary precision in some columns, however the default value could still be float. This would probably be the simplest change.

**Enhance Dtypes Parameter**
Alternatively it would be nice to modify the Dtypes[] parameter in Read_Csv to be a dictionary such that we can specify the particular type we want for a given column name. This will make it a little more robust. (I have customer data with the column names in different orders/ shifted in excel etc. It would be nicer to not have to rely on indexing). The disadvantage here is that we would have to check that column names exist etc.