formatv2 does not keep well-balanced multi-line index expressions

[mvdan]

exec cue-v0.16.1 fmt --diff in.cue
exec cue-v0.17.0-alpha.3 fmt --diff in.cue
exec cue-master fmt --diff in.cue

-- in.cue --
idx: bar[
	"baz"
]

> exec cue-v0.16.1 fmt --diff in.cue
[stderr]
expected ']', found newline:
    ./in.cue:2:7
missing ',' in struct literal:
    ./in.cue:3:1
[exit status 1]
FAIL: repro.txtar:1: unexpected command failure
> exec cue-v0.17.0-alpha.3 fmt --diff in.cue
[stdout]
diff in.cue.orig in.cue
--- in.cue.orig
+++ in.cue
@@ -1,3 +1,3 @@
 idx: bar[
-	"baz"
-]
+		"baz"
+		]

> exec cue-master fmt --diff in.cue
[stdout]
diff in.cue.orig in.cue
--- in.cue.orig
+++ in.cue
@@ -1,3 +1,2 @@
 idx: bar[
-	"baz"
-]
+	"baz"]

Support for a comma was added in v0.17, so ignore the fact that v0.16.0 fails.

Also ignore the fact that v0.17 leads to bad indentation. That's just the old formatter being buggy.

Regardless, formatv2 should be a no-op here; it shouldn't force ] to be collapsed when it's on a new line and well balanced with [ being at the end of a line.

[cuematthew]

Just some notes on this: Need to get a clear indication that the output is for language version v0.17 or better; if we're formatting a pkg from a module less than this we presumably need to ensure the syntax is parsable by eg v0.16 so we need to avoid the \n because of the auto-comma-insert rule.

[mvdan]

Correct, you can rely on the current module's language version, which should be passed through to the formatter via the options API.

One quirk: we assume that the lack of a CUE module means "current language version", i.e. v0.18.0 as of master. So this could mean that a standalone use of cue fmt can silently break older CUE versions that don't know how to parse this. I think this is okay.