Explicitly require csv for for Ruby head

[Mangara]

Some bundled gems are being removed from Ruby and need to be specified in the Gemfile with Ruby 3.4: https://bugs.ruby-lang.org/issues/20187

This PR adds the CSV gem to our Gemfile for use in testing.

I don't think we need to add it to the Gemspec? If users don't use the CSVEnumerator, they shouldn't need the gem, right?

[sambostock]

I don't think we need to add it to the Gemspec? If users don't use the CSVEnumerator, they shouldn't need the gem, right?

My gut feeling for this is that if we want to just work without csv, as long as you don't try to use it, then we'll need something more "clever" to explicitly deal with csv being missing.

While looking into that, I noticed that we actually don't require 'csv' anywhere in the app (we do it in test/unit/csv_enumerator_test.rb, but not lib), despite referencing the CSV constant, which I think is actually a bug (although in practice it's unlikely anyone has run into it, as I'll bet something in most host apps requires it).

I think one approach would be something like

# lib/job-iteration/csv_enumerator.rb

begin
  require "csv"
rescue LoadError
  return
end

module JobIteration
  class CsvEnumerator
    # ...

and an entry in the README about it.

[Mangara]

We only use the CSV constant to check that csv.instance_of?(CSV). Should we instead change that to csv.class.name == "CSV"?

That would make

1. People trying to call the enumerator without a CSV object get the error

CsvEnumerator.new takes CSV object (ArgumentError)

2. People (after the first error?) trying to call the enumerator with a CSV object get the error:

uninitialized constant CSV (NameError)

originating from their own code, which should hopefully get them to require "csv" pretty quickly (even Googling it).

The class would still get loaded for apps without the csv gem, but that seems very minor.