fix expression vocab memory management and incorrect reallocation of aggregates storage

[abalabin-bamfunds]

also fix and add ability to create tables with index to the sys_info example

This PR fixes a problem when having a string-valued expression column on views of tables that frequently update made perspective grow unlimited amount of memory and slow down dramatically with time. While I could not get a test together that would show this clearly, the results of manual testing are below.

Using the updated sys_info example and the JSON pasted below for the "Generated Data" view the observations before the fix looked like this:

after the fix:

JSON:

{
  "version": "4.4.1",
  "columns_config": {},
  "plugin": "Datagrid",
  "plugin_config": {
    "columns": {},
    "scroll_lock": false,
    "edit_mode": "READ_ONLY"
  },
  "settings": true,
  "table": "superstore_4",
  "theme": "Pro Light",
  "title": "Generated Data",
  "group_by": [
    "String 1"
  ],
  "split_by": [],
  "sort": [],
  "filter": [],
  "group_rollup_mode": "rollup",
  "expressions": {
    "S": "concat(\"String 1\", \"String 2\")"
  },
  "columns": [
    "S",
    "Float 2",
    "Float 3",
    "Float 4",
    "Integer 0",
    "Integer 1",
    "Datetime 4",
    "Boolean 0"
  ],
  "aggregates": {
    "Datetime 4": "dominant",
    "S": "any",
    "Boolean 0": "dominant"
  }
}

[texodus]

Thanks for the PR and research, @abalabin-bamfunds !

It's a real issue and the fix and prose are super helpful.

However this fix has a memory safety issue. t_expression_vocab is a wrapper around t_vocab exclusively used to prevent ExprTK from interning new strings while it holds references to old ones, hence the mechanism in this file which calls allocate_new_vocab() instead of t_vocab::extend, the latter of which invalidates existing string references.

Small side issue - the DCO Bot failed your PR for lacking a sig. I can see in the comments that you signed it, but the DCO Bot is not recognizing your sig. Perhaps because it is indented? We really appreciate you making sure to cover this step up front (it means you took the time to read the CONTRIBUTING.md! Thank you!) but the bot can be finicky sometimes, if you did not use -s when committing it should have formatted this correctly. If you did this already, perhaps the DCO Bot needs a PR!

Back to the issue at hand. I've gone ahead and re-implemented a fix and added a test for this issue in #3177 that is memory stable, so closing this PR.

[abalabin-bamfunds]

Thanks @texodus

Agree re memory safety - I did not spot that letting vocab growing over pre-allocated size invalidates pointers and weirdly (and thankfully!) did not get any crashes while running with this fix for a few months. It is cool to have a proper fix.

Re DCO - I think this was a squash commit from pre-signed commits done with -s so I would guess indentation comes from squashing commit messages. Will keep an eye on it, or maybe the DCO bot could allow some whitespace in front on sign-offs?