Fix Noto font bug

A little bit of context.
While working with pdfkit we found this error.
foliojs/pdfkit#1106
We narrowed down the repro to fontkit with this little script
```
const fs = require('fs');
const notosans = require('fontkit').openSync('./NotoSansCJKsc-Bold.otf');
const subsetFor = (text, file) => {
  const run = notosans.layout(text);
  const subset = notosans.createSubset();
  run.glyphs.forEach((glyph) => subset.includeGlyph(glyph));
  subset.encodeStream().pipe(fs.createWriteStream(file));
};
subsetFor('间。楼', 'output.ttf');
```
With this script you can see that glyphs were missing on the subset font if you opened the font with something like fontforge

Regarding the logic behind this fix, when `!used_fds[fd]` was false, last items in `topDict.FDArray` and `used_subrs` weren't really the ones corresponding to the current `fd`. So we are saving a dictionary to find the real index were those values were stored.

What do you think about this fix?

src/subset/CFFSubset.js

@@ -55,6 +55,7 @@ export default class CFFSubset extends Subset {
 
     let used_fds = {};
     let used_subrs = [];
+    let fd_select = {};
     for (let gid of this.glyphs) {
       let fd = this.cff.fdForGlyph(gid);
       if (fd == null) {
@@ -64,15 +65,16 @@ export default class CFFSubset extends Subset {
       if (!used_fds[fd]) {
         topDict.FDArray.push(Object.assign({}, this.cff.topDict.FDArray[fd]));
         used_subrs.push({});
+        fd_select[fd] = topDict.FDArray.length - 1;
       }
 
       used_fds[fd] = true;
-      topDict.FDSelect.fds.push(topDict.FDArray.length - 1);
+      topDict.FDSelect.fds.push(fd_select[fd]);
 
       let glyph = this.font.getGlyph(gid);
       let path = glyph.path; // this causes the glyph to be parsed
       for (let subr in glyph._usedSubrs) {
-        used_subrs[used_subrs.length - 1][subr] = true;
+        used_subrs[fd_select[fd]][subr] = true;
       }
     }
 

test/subset.js

@@ -125,5 +125,27 @@ describe('font subsetting', function() {
         return done();
       }));
     });
+
+    it('should produce a subset with asian punctuation corretly', function(done) {
+      const koreanFont = fontkit.openSync(__dirname + '/data/NotoSansCJK/NotoSansCJKkr-Regular.otf');
+      const subset = koreanFont.createSubset();
+      const iterable = koreanFont.glyphsForString('a。d');
+      for (let i = 0; i < iterable.length; i++) {
+        const  glyph = iterable[i];
+        subset.includeGlyph(glyph);
+      }
+
+      return subset.encodeStream().pipe(concat(function(buf) {
+        const stream = new r.DecodeStream(buf);
+        const cff = new CFFFont(stream);
+        let glyph = new CFFGlyph(1, [], { stream, 'CFF ': cff });
+        assert.equal(glyph.path.toSVG(), koreanFont.glyphsForString('a')[0].path.toSVG());
+        glyph = new CFFGlyph(2, [], { stream, 'CFF ': cff });
+        assert.equal(glyph.path.toSVG(), koreanFont.glyphsForString('。')[0].path.toSVG());
+        glyph = new CFFGlyph(3, [], { stream, 'CFF ': cff });
+        assert.equal(glyph.path.toSVG(), koreanFont.glyphsForString('d')[0].path.toSVG());
+        return done();
+      }));
+    });
   });
 });