fix(help): spell Arguments-block names the way the usage line does

[dspangen]

omni models merge-branch --help described its two positionals twice, under two different names:

Arguments:
  <modelid>     Model UUID
  <branchname>  Branch name

Usage:
  omni models merge-branch <model-id> <branch-name> [flags]

The Arguments block exists to tell a reader that the second positional is a branch name, not a UUID. That only lands if both blocks are recognizably about the same argument — under two spellings they read as four arguments.

The usage line has used canonicalName since #77; argumentsHelp kept calling slugify, which lowercases but doesn't split camelCase. Both now use canonicalName.

Help text only — nothing is parsed from these strings, and flag/arg matching is unaffected.

Details: scope and verification

Scope

One line in argumentsHelp. Only the path-param branch needed it; body-shorthand arg names (model-id, prompt, …) are authored in kebab-case already and were never affected.

Every generated command with a camelCase path param was showing the wrong spelling — ai conversation-detail (<conversationid>), ai job-cancel (<jobid>), models merge-branch, and so on.

After

Arguments:
  <model-id>     Model UUID
  <branch-name>  Branch name

Usage:
  omni models merge-branch <model-id> <branch-name> [flags]

Verification

make build && go test ./... — all packages pass.

TestBuildCommand_ArgumentsSection updated to expect the canonical spelling, and a new TestGenerateCommands_ArgumentNamesMatchUsageLine walks every generated command in api/openapi.json and asserts each <placeholder> in the usage line appears in the Arguments block — the two are built by separate code paths, which is how they drifted in the first place.

Both fail against the pre-fix code (checked by reverting the one-line change): the spec-wide test flags ai conversation-detail, ai job-cancel, and the rest.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UP53YBDjhQkkn3WAqMR3sf