topic: improve the revset used to return name->nodes mapping The 'topic' version already filters public changeset out, and skipping 'changectx' creation will helps performance.

topic: update the changectx method to respect phases The topic is no longer returned once the changeset is no longer mutable (unless explicitly requested).

topic: avoid crash when topic is loaded but not enabled for a repository

compat: update test to match core default state This output changed in 560ceb654180.

topicmap: massive rework Massively rework the way we build and use topicmap. This bring massive performance benefit. Topic map use to be a fully independant thing that we would switch on and off globaly. The caching on disk was broken so the performance were atrocious. Intead, now the topic are inherited from the 'immutable' map. We gave up on storing them on disk for now since the mutable set is usually small enough. The activation is done by hacking new "filter" on the repository and detection when they are one. This is hacky but core is hard to wrap here. Overall this whole wrapping is really scary and we should massage core API to help it.

topic: use the 'topic' revset in namespace This revset performance just got improved.

topic-revset: update the revset to no longer build changectx Atribute access is expensive, and changectx are very expensive to build, so we skip these for better performance. Before: ! wall 0.012195 comb 0.020000 user 0.020000 sys 0.000000 (best of 217) After: ! wall 0.008816 comb 0.010000 user 0.010000 sys 0.000000 (best of 303) (Before this changeset parent) ! wall 0.213261 comb 0.210000 user 0.210000 sys 0.000000 (best of 45)

topic-revset: changectx creation in the revset The current mutability filter is very expensive, we use a more direct and efficent way to do so: Before: ! wall 0.213261 comb 0.210000 user 0.210000 sys 0.000000 (best of 45) After: ! wall 0.012195 comb 0.020000 user 0.020000 sys 0.000000 (best of 217)

topics: add some noise to rewrittent changeset to prevent hash cycle If we have a changeset with topic `x`, we change it's topic to `y`, fine. When we change it's topic back again to `x`, we get the hash of the the obsoleted changeset which had the topic `x` initially. The same happens for few more cases like clearing the topic of a changeset which initially had no topic. This approach is influenced from cmdutil.amend and other commands (rebase, histedit, etc…)

topic-change: update the working copy along when changing topic of '.' This avoids leaving the working copy behind, removing another large parts of issue5441.