Sometimes we receive bufio.Scanner: token too long is the parsed
string is too long

Let's increase the buffer size

RAM usage of parsing LSIF references are now back to normalcy and
feature works.

This removes the code in workhorse for determining whether parsing
references is enabled or not. It'll now always be enabled.

In order to be able to "Find references" of a given range or
definition, we need to be able to read `references` items from
the LSIF file.

This will then be written to the JSON file for each range via
`references` property.

The property will look something like:

```
"references": [
  {
    "path": "main.go#L7"
  }
]
```

Each reference will be an object with `path` property. The `path`
property will point to the exact line where it is being used.

This is currently behind a feature flag (`code_navigation_references`)
which will set `ProcessLsifReferences` header to `true` when
enabled. If `false`, the `references` won't be read and generated.

Define Id type with custom marshalling which accepts either
int or string, because some languages (for example, Javascript)
define ids in LSIF as ints, but some languages (like Golang) use
string type for ids

It also changes int to uint32 for easier binary conversion

LSIF document is converted into multiple JSON files and archived
as a ZIP file

https://gitlab.com/gitlab-org/gitlab/-/issues/212384