support multi-command statements

[pav-kv]

The amount of boilerplate to run a sequence of commands in a datadriven test is excessive. For example, here is a typical script. This is visually unsatisfying (the amount of signal per screen is low, and one has to scroll to read it), and slow to type.

It would be good to be able to batch multiple commands in a single statement, i.e. instead of:

command1
----
output1

command2
----
output2

command3
----
output3

have something like:

command1
command2
command3
----
output1
output2
output3

Variations are possible. For example, multiple ok outputs could be squashed into a single ok:

command1
command2
command3
----
ok

This format is in conflict with the current syntax which treats a multi-line statement as a single command with additional input.

We may need some way to mark the multi-command statements. For example:

+ command1  # assumming there are no existing tests with "+" command
+ command2  # we could restrict command names to be "identifiers" starting with a letter
+ command3
----
output1
output2
output3

or:

---- # or other special preamble
command1
command2
command3
----
output1
output2
output3

The first option works better with multiple multi-line commands:

+ command1
<input> # must not have a line starting with "+"
+ command2
<input>
+ command3
<input>
----
output1
output2
output3

Although the given syntax could be restricted to single-line commands, for simplicity. This would give most of the value anyway.

For extra niceness, it would be good to support inline comments next to the commands. With all the modifications, the scripts can look more satisfactory:

Details

# Start with 3 nodes, with a limited in-flight capacity.
add-nodes 3 voters=(1,2,3) index=10 inflight=3
----
ok

+ campaign 1  # elect node 1 to be the leader
+ stabilize
----
ok

+ propose 1 prop_1_12  # propose a few entries
+ propose 1 prop_1_13
+ propose 1 prop_1_14
+ process-ready 1  # store entries and send proposals to followers
----
ok

# Expect that in-flight tracking to nodes 2 and 3 is saturated.
+ log-level debug
+ status 1
+ log-level none
----
1: StateReplicate match=14 next=15
2: StateReplicate match=11 next=15 paused inflight=3[full]
3: StateReplicate match=11 next=15 paused inflight=3[full]

# etc

[pav-kv]

If the use of multi-command statements is expected to be high, we could make the single-line statements the default with some kind of "pragma" at the top of the file, and then default to not needing to type the "+ " prefix for all the commands. Instead, now the multi-line commands would need to have a prefix:

#! single-line-commands  <-- special command that flips the default syntax
#! version=2  <-- or we could just say that it's the next version of the syntax

# or we could make a datadriven/v2 release, and not need extra boilerplate to
# change the syntax. I would prefer this approach. In v2, though, we would add
# support for "#!" pragmas like the above, if minor options/annotations need to
# be introduced without bumping the major Go package version.

command1
command2
command3
----
outputs

$ multi-line-command
<input>
----
ok

[pav-kv]

After some discussion with @RaduBerinde: we should not complicate the syntax. Instead, we can export helpers for parsing the input of a command as if it's a list of datadriven commands, to achieve the same effect.