refactor(coderd/x/chatd)!: drop the context-limit fallback for old compaction summaries - #31002

This PR breaks compatibility with compaction summaries written before #28568, which no release includes yet, so summaries from released versions have this shape. Their chat_summarized tool result has context_limit_tokens (then the trigger's window) and no trigger_context_limit_tokens. The check for whether a summary was recorded under another trigger used to fall back to context_limit_tokens for those results. The PR deletes that fallback and its noTriggerContextLimit sentinel, so the check reads only threshold_percent and trigger_context_limit_tokens. Coder Agents is experimental (coderd/x), and our rule for experimental persisted data is to drop the legacy shape outright and delete the helpers that existed only to parse it, instead of carrying a migration.

What old chats do now: a missing trigger_context_limit_tokens defaults to the current trigger, the same default an empty result already gets, so an old summary compares only on its threshold. If the trigger window has not changed since that summary, nothing changes. If it has (for example, a compaction model override now binds) and the first assistant reply after the summary is over the current threshold, the chat no longer compacts again. The turn ends with "Conversation compaction could not reduce the history below your compaction threshold" instead. A manual compaction, a different threshold or a new conversation clears it.

Why it is safe

• The compaction writer always records trigger_context_limit_tokens. I mutated the remaining comparison and its defaults and ran ./coderd/x/chatd/ with go test -overlay. The tests catch all six mutants:
  • Trigger limit or threshold not compared: TestCompactionStatusFromHistory/summary_under_a_different_trigger_compacts_again fails.
  • Default for the trigger limit or the threshold removed: the still over limit cases that use an empty {} result fail.
  • Comparison against the chat window instead of the trigger: that case and TestActiveServer_CompactionModelOverride fail.
  • Writer omitting trigger_context_limit_tokens: TestActiveServer_Compaction and others fail.
• The deleted case "summary without trigger limit compares context limit" covered only the old result shape. git grep noTriggerContextLimit finds nothing. site/ still reads context_limit_tokens, which keeps its meaning (the chat model's window).

Review focus

The comment kept in summaryRecordedUnderOtherTrigger describes the empty-result defaults, which stay.

Validation

• go test ./coderd/x/chatd/ -count=1 -p 4 -parallel 4: ok
• go vet and golangci-lint run --fix=false ./coderd/x/chatd/...: clean; pre-commit hook: passed

Stack

This stack retires chatd compatibility paths for chat data that older Coder versions stored.

1. test(coderd): stamp chat message fixtures with the current content version #30996 stamp chat message fixtures with the current content version
2. test(coderd/x/chatd/chatprompt): drop tests of the legacy content codec #30997 drop tests of the legacy content codec
3. refactor(coderd)!: retire the V0 chat message content decoder #30998 retire the V0 chat message content decoder
4. refactor(coderd/x/chatd): inline the model config provider identity #31000 inline the model config provider identity
5. refactor(coderd/x/chatd)!: drop the provenance fallback for unstamped OpenAI reasoning #31001 drop the provenance fallback for unstamped OpenAI reasoning
6. refactor(coderd/x/chatd)!: drop the context-limit fallback for old compaction summaries #31002 drop the context-limit fallback for old compaction summaries (this PR)
7. refactor(coderd/x/chatd)!: retire the close_agent tool alias #31003 retire the close_agent tool alias

Xum, an AI agent, opened this PR for @ibetitsmike and wrote this description.