Fix and improve float formatting

[fmeum]

12e+34 is no longer broken into 12e + 34.

Also adds a rewrite pass that turns e.g. .23E+021 into 0.23e21:

• adds a leading 0 to floats
• removes leading +s and 0s from the exponent
• lowercases E

[vladmos]

Do I understand it correctly that it allows some invalid expressions like 1e2e3 or 1.2e3.4? Not insisting that it should be fixed here, maybe it's actually better to format the rest than fail.

[fmeum]

I added a rewrite pass that adds a leading 0, removes leading 0s and +s from the exponent and lowercases the E.

[vladmos]

Shouldn't this E also be lowercased?

[vladmos]

Or this is invalid and just left as is?

[fmeum]

This is not a float, but a hex number. We could discuss whether hex literals should be canonicalized in some way, but I would prefer to handle that in a separate PR.