prog: humanize syzlang in seed programs and reproducers

[a-nogikh]

https://github.com/google/syzkaller/blob/master/docs/program_syntax.md

Some relatively simple steps we could do to make our seed programs much easier to write.

1. Support multi-line call statements. If we define some complex variables in normal programming languages, arrays and fields are often split into several lines to aid comprehension.
   (As discussed with @dvyukov, we could implement it very easily by restricting the set of the tokens before the line break. {, [ and , seem to be enough for almost all cases.)

2. Add optional field names for structures.
   We probably still don't want structure names because in the presence of templates they may be vey unexpected-looking

3. Auto-rewrite the existing seeds using 1 and 2. It could also be some auto-formatter included in make format like we do for the descriptions in sys/*/*.

4. Add support for referencing const values by name.

[a-nogikh]

A potential problem in case we support multi-line calls: it becomes much harder to parse out programs from the noise.

syzkaller/prog/parse.go

Line 20 in e12e5ba

func (target *Target) ParseLog(data []byte, mode DeserializeMode) []*LogEntry {

In ParseLog, we currently to append each subsequent line and verify if the program can still be parsed. It won't work for multi-line calls since we have no decision until we have appended the last line of the call, which potentially exponentially increases the search space.

There might be some heuristics that would work reasonably well though.

[dvyukov]

In ParseLog, we currently to append each subsequent line and verify if the program can still be parsed.

Now it should be less of a problem since we collect programs in manager and write them out to the log without corruptions.
The careful parsing logic was required when syz-fuzzer was logging programs intermixed with kernel output.

[dvyukov]

We can also log w/o using multi-line more.