Delete code pertaining to incremental compilation

[rsheeter]

Rename --incremental to --emit-ir; delete all code pertaining to incremental operation, simplify the IR Source trait as a result. Net reduction of ~2k loc, simpler execution path, and less work done for a clean compile.

For context, the incremental code was written prior to understanding just how dynamic the graph of tasks can be. We default it to off and it's not clear its completely reliable. If we find we want incremental one day we can start fresh; until then it's much simpler to just not do all the associated bookkeeping.

Apologies for the huge diff, it's almost all just deleting bookkeeping.

My intuition is this will lead to further simplifications.

We're doing less work so it shoudl be faster, but is it? - timed on M1 MacBook it appears it is:

# fontmake
$ (cd ~/oss/googlesans && rm -rf variable_ttf && time fontmake -o variable -- source/GoogleSans/GoogleSans.designspace)
real	3m4.600s

# fontc, main vs nocs branch
$ hyperfine \
  --parameter-list branch main,nocs \
  --setup "git checkout {branch} && cargo build --release" \
   --warmup 5 --runs 20 \
   --prepare 'rm -rf build/' \
   "target/release/fontc --source ../googlesans/source/GoogleSans/GoogleSans.designspace"

Benchmark 1: target/release/fontc --source ../googlesans/source/GoogleSans/GoogleSans.designspace (branch = main)
  Time (mean ± σ):      4.137 s ±  0.032 s    [User: 7.495 s, System: 7.130 s]
  Range (min … max):    4.068 s …  4.180 s    20 runs
 
Benchmark 2: target/release/fontc --source ../googlesans/source/GoogleSans/GoogleSans.designspace (branch = nocs)
  Time (mean ± σ):      3.465 s ±  0.041 s    [User: 7.351 s, System: 6.561 s]
  Range (min … max):    3.348 s …  3.529 s    20 runs
 
# THE IMPORTANT PART
Summary
  target/release/fontc --source ../googlesans/source/GoogleSans/GoogleSans.designspace (branch = nocs) ran
    1.19 ± 0.02 times faster than target/release/fontc --source ../googlesans/source/GoogleSans/GoogleSans.designspace (branch = main)

# That is, on main we build GS ~45x faster than fontmake and on this branch we build it ~53x faster.

JMM if happy.

[cmyr]

Assuming no performance impact I think this will be a huge win, and I share the intuition that getting rid of it will unlock a range of other improvements.

[rsheeter]

Oh I maybe worded that poorly, if there is a perf impact I expect it to be positive.

[cmyr]

just did a skim, looks very promising!

[cmyr]

looks good, very happy to see this merged!