Document, clean and stabilise phragmen's behavior with dupe candiadtes and voters.

[kianenigma]

Both staking and elections should not allow duplicate candidates, but still, it happened #4592. A simple fix should make sure that phragmen is internally also dealing with this, in the proper way. Currently, out of luck, the dupes are being ignored, but it should be done in a more mature manner.

[shawntabrizi]

@kianenigma lets do this together this week

[kianenigma]

More precisely:

• dupe candidate
• dupe voter
• dupe target

[kianenigma]

I have found a bug related to this: using offline-phragmen, I realized that this assertion in the reduce code is failing, only in polkadot at the moment:

substrate/primitives/npos-elections/src/reduce.rs

Line 172 in 252b146

debug_assert!(other_votes_count <= 2);

Which means:

Someone submitted a duplicate vote, and it somehow made its way into the reduce code. After tweaking the error message, I find this dude:

Surprising is that it seems like seq-phragmen elected both of the duplicate edges. This should have not happened, and needs investigating.

[kianenigma]

More evidence:

#17 --> NO_IDENT [8889bb12ffc22c93e6190aeac259184d7181bed3f0cc9938d27315f8e61c8c4a (1462Rm6Z...)] [total backing = 16934,734DOT (16,934,734,626,230,101) (4 voters)] [own backing = 0,010DOT (10,000,000,000)]
  Voters:
    #1 [amount = 6720,128DOT (6,720,128,027,299,203)] 75d7a967cb385d6f867d150f0727d9cba858c65d192530b44298a06c9bfdea9c (13fWfiEQ...)
    #2 [amount = 0,000DOT (42,560,803)] 75d7a967cb385d6f867d150f0727d9cba858c65d192530b44298a06c9bfdea9c (13fWfiEQ...)
    #3 [amount = 10214,596DOT (10,214,596,556,370,095)] aa6908fa06b85012d44fc8966d9760fb0b0fd9282be9f239927a0089e29e94c8 (14rSLrbz...)
    *#4 [amount = 0,010DOT (10,000,000,000)] 8889bb12ffc22c93e6190aeac259184d7181bed3f0cc9938d27315f8e61c8c4a (1462Rm6Z...)

[kianenigma]

Albeit, on chain this seems to be fine, meaning that the final exposure is sane, yet it might be dubious and unfair.

[kianenigma]

As an additional part of this, Both staking and elections-phragmen should ensure that the votes submitted by the voter are all unique, rather than relying on the underlying election crate to filter it out for them.

[coriolinus]

One random note unrelated to this PR per se: this pallet had quite a bad rap for not having a well defined behaviour with duplicate votes and candidates and targets etc. Now, finally, everything that goes through setup_input (called prior to seq_phragmen and phragmms) gets sanitized. Would be good to ensure these functions have the same standard.

Originally posted by @kianenigma in #8236 (comment)