CI takes 28 min. Shrink via (1) download vyper 1x (2) parallelize github actions

[trentmc]

Background / motivation

Total CI runtime is 28 min. Here's an example log from a successful end-to-end run.

From a conversation between Trent & Alex, based on inspecting the log, we see a couple quick wins to speed this up.

• Issue 1: downloads vyper three times
• Issue 2: only after barge download completes does it compile. Compile takes real time too

TODOs:

Optimize github actions as follows.

• For issue 1: download Vyper just once
• For issue 2: (a) start barge download, then (b) download vyper & compile in parallel. Once both (a) and (b) are done, then do the rest.

Detail: Downloads vyper three times

This happens in the "Test with Brownie" github action.

Downloading from https://github.com/vyperlang/vyper/releases/download/v0.3.1/vyper.0.3.1.linux
  0%|          | 0.00/17.2M [00:00<?, ?iB/s]
  3%|▎         | 558k/17.2M [00:00<00:02, 5.58MiB/s]
...
100%|██████████| 17.2M/17.2M [01:10<00:00, 245kiB/s]
Using vyper version 0.3.1
...
...
100%|██████████| 17.2M/17.2M [01:10<00:00, 245kiB/s]
Using vyper version 0.3.1
vyper 0.3.1 successfully installed at: /home/runner/.vvm/vyper-0.3.1
Downloading from https://github.com/vyperlang/vyper/releases/download/v0.2.7/vyper.0.2.7+commit.0b3f3b3.linux

  0%|          | 0.00/20.6M [00:00<?, ?iB/s]
  3%|▎         | 519k/20.6M [00:00<00:03, 5.19MiB/s]
...
100%|██████████| 20.6M/20.6M [01:02<00:00, 328kiB/s] 
vyper 0.2.7 successfully installed at: /home/runner/.vvm/vyper-0.2.7
Downloading from https://github.com/vyperlang/vyper/releases/download/v0.2.4/vyper.0.2.4+commit.7949850.linux

  0%|          | 0.00/20.6M [00:00<?, ?iB/s]
  3%|▎         | 639k/20.6M [00:00<00:03, 6.39MiB/s]
...
100%|██████████| 20.6M/20.6M [00:43<00:00, 470kiB/s]
vyper 0.2.4 successfully installed at: /home/runner/.vvm/vyper-0.2.4
Downloading from https://solc-bin.ethereum.org/linux-amd64/solc-linux-amd64-v0.8.12+commit.f00d7308
Compiling contracts...

[trentmc]

This is relatively high priority because doing it will greatly reduce CI runtime, for faster turnaround of df-py in general