It is clear that the original intent was to have the data available as part of the data feed available at that point, it is also clear that nothing has been done to merge these two datasets, so lets do that now.

My preference in this case is to use forEach to iterate over countries and modify each entry to include the entire matching record from the csv file. You could use map but in this case we are already loading heaps of data into memory, we don't need to retain the unchanged version of the countries JSON as well as the new output that map would have produced.

The other reason that I recommend injecting the data from the CSV into the array of countries is that we can avoid the overhead of a lookup when we need to access the data. Arranging the csv data and the country data into arrays of the same length and sorted so that the same indexes correlate to the same country in each file incurs the same over head as injecting the data and makes it more deterministic. It is also easier to support scenarios where one array has more values that the other.

All we need to know is the property mapping to identity the unique record in each file. We can see that the csv file is using 2 letter country codes:

I have picked two pairs Australia, Austria and India, Indonesia for comparison as they cover common matching issues, Most countries have a 2 letter postal code, and it is common for people to pick this first, however India's postal code is 3 letters, Indonesia's is 4 and Austria have 1. I find this dataset particularly interesting for how close the name terms are lexicographically yet how different their unique codes are, so I always check with these specific cases first before running a test or proof across an entire dataset.

The feature property metadata in the json file has a lot of information, importantly, it also has the ISO_A2 value, which is the ISO standard 2 letter country code.

there are lots of different ways to do this, I'm just going to find the matching record and stuff it into a new property called csv using a verbose method to make it easy to follow:

// Convert the TopoJSON to GeoJSON
countries = {
  const geo = topojson.feature(map_layer, map_layer.objects.countries);
  geo.features.forEach(feature => {
    if(feature.geometry) {
      feature.centroid = centroid(feature);
    }
    let code = feature.properties.ISO_A2;
    let cData = country_data.find(x => x.CountryCode === code);
    feature.properties.csv = cData ?? {
  areasqkm: "",
  population: "",
  airports: "",
  gdpgrowthrate: "",
  inflationrate: "",
  unemploymentrate: "",
  popnbelowpoverty: "",
  medianage: ""
};
    return feature;
  });
  return geo;
}

Now the data is available, including a null record to simplify error handling later. Lets add this to the tooltip content first, to verify that it works, i'll dump all the extended data just so we can explore the different values:

<div>
  <h3>${hover.NAME}</h3>
  <div style="margin: 10px">
    <h4>CountryCode : ${hover.ISO_A2}  </h4> 
    <h4> ContinentCode : ${hover.CONTINENT}  </h>
    <h4>${parameter } : ${d3.format(",")(hover.csv[p])}</h4>
    <hr style="padding:0px"/>
    <smaller>
     Area Sqkm: ${d3.format(",")(hover.csv['areasqkm'])}<br/>
     Population: ${d3.format(",")(hover.csv['population'])}<br/>
     Airports: ${hover.csv['airports']}<br/>
     GDP Growth Rate: ${d3.format(",")(hover.csv['gdpgrowthrate'])}<br/>
     Inflation Rate: ${d3.format(",")(hover.csv['inflationrate'])}<br/>
     Unemployment Rate: ${d3.format(",")(hover.csv['unemploymentrate'])}<br/>
     Pop. Below Poverty: ${d3.format(",")(hover.csv['popnbelowpoverty'])}<br/>
     Median Age: ${hover.csv['medianage']}<br/>
    <smaller>
  </div>
</div>

Now you just need to apply those property mapping paths to your bubble logic, I'm actually going to leave that logic out here, the main issue was how to make the data available, there are practicality issues to using the direct value to render proportional shapes. Very quickly you end up with some items so small you can't click or see them, other shapes will have such large discrepancies that functions like forceCollide() will just make a mess... have a look at zimbabwe's inflation rate in this bubble map:

It is often more practical to map values to specific size domains, a scale of 1-5 is usually enough but you can go to 10, where 1 is the smallest on the map, and 10 the largest, then trim the data range into standard deviations so that outliers below all get 1 and outliers above all get the same max.

The techniques and reasoning are well outside of this discussion space.