Added support for "default" colors in ScreenBuffer

[samizdatco]

Since all but one of the 32 attr bits were already in use, I expanded the attr size from 4 bytes to 6 and in the process reordered the 'colors' and 'flags' such that the colors are now in the upper 16 bits and the flags in the lower 16 (which ends up being more convenient for checking bitmasks since the bitwise operators all truncate Number values to 32 bits).

After adding this extra space to the middle of the attr field I assigned two new flag bits (at 17 & 18) that indicate whether a color and/or bgColor have been selected respectively. If they were left undefined (or passed a string value of "default"), the flag bits are unset. The escape-generating methods now consult these flags to decide whether colors should be set based on the color bits in attr or the terminal's 'default' fg/bg colors should be used.

Some Caveats

• I added 16 bits to the attr size but am only using two of them at the moment so it's a bit on the wasteful side. Unless you can imagine a use for the rest of that space, it might make more sense to set ScreenBuffer.prototype.ATTR_SIZE to 5 rather than 6. I believe that the rest of the code should adapt without error if you just change this one value.

• Since the generateDeltaEscapeSequence() method may need to set the fg and bg colors back to the 'default' value independently I added the defaultColor and bgDefaultColor escapes to term.optimized.

• I incremented the version number on .sbuf files to 3 but ended up modifying the load- and saveSyncV2() methods to branch based on the version number in the header rather than creating brand new (and nearly identical) V3 methods. If keeping the historical serialization formats in the source is important to you, you might want to paste the old versions back in.

• I used the demo/spaceship.js script as my main test-case and made some minor edits to it in places where it was relying on a black bgColor being present by default.

• I wasn't quite sure if the object2attr and attr2object definitions were repeated intentionally or not (since the implementations appeared identical). For now I'm just assigning the same function to both the 'class' and 'instance' methods...

[cronvel]

@samizdatco Like I said in the related issue: thanks for your contribution, but unfortunately I was already working on that.
I'm trying to avoid a SemVer Major change and aim for a backward compatible change.

I have cherry-picked "delete previous/next word keybindings" and pushed version 1.9.0 with your change to .inputField(). Thanks! ;)

[cronvel]

@samizdatco I've implemented it. It landed on v1.10.0.
I've done it so it does not break the compatibility.
I removed an unimplemented flag that was reserved for further usage, but I think that this flag would be equivalent to some combination of existing flags (mainly transparency flags).