Small optimisations to PYON encoding

[SquidDev]

We've been seeing a couple of performance issues when serialising large PYON files¹. This PR makes a couple of changes to the PYON encoder to improve performance without changing behaviour:

• Use json.dumps to quote strings, rather than str.translate, which is significantly faster. This does mean that unicode characters are encoded as escape sequences instead, but they should serialise the same.
• Build up a list of strings which are concatenated at the end, rather than concatenating as we go along. This avoids the quadratic complexity of string concatenation.

Here is a quick summary of the time taken to serialise a 1.7GiB PYON object across the commits in this PR:

Commit	Time (s)
c0a7ed3	97.9
0313881	14.3
be74f9f	7.7

Footnotes

1. Most notably within the ARTIQ dashboard's StateManager, where serialising the dashboard state can cause the UI to hang for several seconds. ↩