Slim down the manifest.json artifact

[drewbanin]

Describe the feature

We've seen some pretty big manifests in the wild. The size of this file is necessarily going to scale linearly with the number of objects in your project + lines of code, but if there are obvious things we should do to slim down this file size, we should do them.

Ideas:

• remove duplicated/unused sql statements (raw/injected/wrapped - do we always need all of these? or are some no longer sensible?)
• remove file_contents from docs nodes
• remove raw_sql from macro nodes

Who will this benefit?

This should make websites (like the docs site) more responsive and faster to load

[beckjake]

remove duplicated/unused sql statements (raw/injected/wrapped - do we always need all of these? or are some no longer sensible?)

#2220 removes wrapped_sql. Given that PR, I think removing injected_sql should be straightforward - we can just overwrite the old compiled_sql instead of setting injected_sql. Whenever that merges I'll rebase on it and get rid of injected as well.

remove file_contents from docs nodes

This was as straightforward as it seemed.

remove raw_sql from macro nodes

Currently we use this in the template caching code. I've remove it from parsed macros by making the cache operate on the macro_sql instead of the full file. I'm not sure what impact that will have on performance in the average dbt project (positive or negative). I think it might hurt error messages, too.

Some other things I thought of:

• We should also consider if we want to flip the omit_none flag on the manifest.
• We should find a way to minimize repetitive config blocks/node fields by omitting empty dicts/lists/objects. post-hooks: [] should just result in an omitted post-hooks config item, etc. Same goes for empty metadata, depends_on, column info, refs, ...
• Seeds don't need sql-related fields
• trim out path vs original_file_path vs root_path vs seed_file_path
  • root_path probably belongs somewhere else entirely in the manifest, in a project-name-to-project-metadata mapping. That would save some repetition!

[beckjake]

Some preliminary work:

|-------------------------------|----------------------|----------------------|
|            desc               |     size (bytes)     | % of octavius-catto  |
|-------------------------------|----------------------|----------------------|
|   dev/octavius-catto          |        867056        |         100          |
|-------------------------------|----------------------|----------------------|
|  No file_contents or raw_sql  |        174359        |         20.1         |
|-------------------------------|----------------------|----------------------|
|   above, plus omit_none=True  |        170837        |         19.7         |
|-------------------------------|----------------------|----------------------|
| no file_contents or raw_sql,  |        172927        |         19.9         |
| rebased to #2220              |                      |                      |
|-------------------------------|----------------------|----------------------|
|   above, plus omit_none=True  |        169531        |         19.6         |
|-------------------------------|----------------------|----------------------|

The size comes from rm -rf target; dbt run && dbt docs generate && ls -l target/manifest.json

I spent a bit of time on the omitting defaults code - it was less helpful than I hoped and involved some breaking changes to how hologram serializes things, so that's probably a dead end.

Significant further changes are probably going to have really diminishing returns.

[drewbanin]

@beckjake awesome! An 80% reduction from removing file_contents and raw_sql sounds like a really awesome start. I have some reservations about using omit_none=True on principal, so it's convenient that it doesn't provide a meaningful lift here :)

Can you open up a PR for this change?