[processor/tailsampling] Replace invert sampled with drop policy

[portertech]

Description

This pull-request replaces the use of InvertSampled and InvertNotSampled decisions with a new policy type to explicitly drop traces regardless of other policy decisions. The new drop policy type behaves much like an and, however, it returns a Dropped decision when all of its policies return Sampled. A Dropped decision takes precedence over all others. Other policy types will either return Sampled or NotSampled. The string, numeric, and boolean filter policies still support invert_match, which continues to flip the decision for the individual policy. Let invert_match be simple.

This is a breaking change, as existing top-level (not within an and or composite) string, numeric, and boolean filter policies using invert_match will no longer have the same impact on the final decision. These policies will need to be replaced with a drop, effectively wrapping the existing policy.

Using an example invert_match policy from the readme:

{
    name: drop-it-like-its-hot,
    type: drop,
    drop: {
        drop_sub_policy: 
        [
            {
                name: team_a-do-not-sample,
                type: boolean_attribute,
                boolean_attribute: { key: app.do_not_sample, value: true, invert_match: true },
            }
        ]
    }
}

This pull-request also optimizes decision policy evaluation. It will make an earlier decision when drop policy is not in use, making a decision on the first Sampled. This reduces tick/decision latency, leading to improved stability and performance.

Related Issues

• [processor/tailsampling] Allow invert matches in composite policy to continue processing #36673
• [processor/tailsampling] invert_match not given precedence when inside and policy #33656
• Tail-based sampling with policies evaluated with first match mechanism #36795
• [processor/tailsampling] numeric attribute invert match evaluation not prioritized #34296
• [processor/tailsampling] invert_match precedence  #34085
• Unexpected behavior of And policy #29637
• Tail-sampling evaluation completing but not filtering out any traces for Grafana Tempo Distributed  #27049
• Probably more 😅

[portertech]

@rjtferreira I am curious to know what you think of this.

[portertech]

@jpkrohling @evan-bradley I would love your input on this 🙏

[jpkrohling]

I'm the first to say that the invert logic is messed up, but unfortunately, it's used in a lot of places. There are a couple of things I'd like to have:

• either as part of this PR or coming before this PR: a set of examples with their expected outcomes. We have some already, but I feel like we need a real database of examples for people to understand the scenarios. Bonus points if we can make them runnable/tested as part of the CI. Having real use-cases and examples for the policies and the implications when applied together will make it easier for people to use and report bugs.
• this PR definitely needs a feature flag, to enable the new behavior without affecting current users. Take a look at our deprecation policy, but in short: current behavior is the default behavior for two more versions, we warn users about the upcoming behavior change on the next version already, we change the default behavior after N+2, and we remove the feature flag and the old behavior in a few month's time.