feat(ourlog): Only enable less destructive PII on the log body

[Dav1dde]

Applies only non-destructive rules to $log.body, also re-uses pii fixtures from existing tests.

[jjbayer]

Shouldn't this still be true? E.g. $logentry.formatted is also in REPLACE_ONLY_SELECTOR but still has pii = true.

[Dav1dde]

$logentry.formatted is explicitly removed from PII instead:

static DATASCRUBBER_IGNORE: LazyLock<SelectorSpec> = LazyLock::new(|| {
    "(debug_meta.** | $frame.filename | $frame.abs_path | $logentry.formatted | $error.value | $request.headers.user-agent)"
        .parse()
        .unwrap()
});

[k-fish]

Is this some weirdness around legacy? I'd expect Pii::Maybe to be advanced rules only, is this because

relay/relay-pii/src/convert.rs

Lines 113 to 122 in bbe038d

	applications.insert(
	REPLACE_ONLY_SELECTOR.clone(),
	vec![
	"@email:replace".to_owned(),
	"@creditcard:replace".to_owned(),
	"@iban:replace".to_owned(),
	"@usssn:replace".to_owned(),
	"@bearer:replace".to_owned(),
	],
	);

is inserting it back as 'advanced' / new PII rules?

[Dav1dde]

Legacy (default scrubbers) are converted to 'advanced rules' there is some magic where these default rules won't apply to maybe. But we opt out in the conversion and inject these extra rules to for specific fields now. There is a longterm plan now to unify this, but until we have that, this is the workaround.

[k-fish]

I think this is fine, however I don't know the ramifications of removing the password:filter. Although it is destructive, it catches auth / secrets / api keys and if those are important we really should try to replace that behaviour if it was previously working for users.

[k-fish]

Is this some weirdness around legacy? I'd expect Pii::Maybe to be advanced rules only, is this because

relay/relay-pii/src/convert.rs

Lines 113 to 122 in bbe038d

	applications.insert(
	REPLACE_ONLY_SELECTOR.clone(),
	vec![
	"@email:replace".to_owned(),
	"@creditcard:replace".to_owned(),
	"@iban:replace".to_owned(),
	"@usssn:replace".to_owned(),
	"@bearer:replace".to_owned(),
	],
	);

is inserting it back as 'advanced' / new PII rules?