Retain chars with semantic or phrasing meaning

[lionel-rowe]

Fixes #7

Manually tested with this input, spoken output is as expected with no errors:

const tts = new EdgeTTS(`1 < 2 & 6 > 5, a\x00b, c\x1fd, foo "bar 'baz'", one (two) three, x\ty\r\nend!`, 'en-GB-SoniaNeural')
new Audio(URL.createObjectURL((await tts.synthesize()).audio)).play()

/regex/v is baseline 2023, depending on minimum support requirements you might want to swap it out for a hard-coded list of control chars a-la the version in src/utils.ts.

[travisvn]

I responded to the premise of this PR in this comment #7 (comment)

Posting here for visibility