feat: model /template/update fields as nullable so channels can be patched

[ysyneu]

What

POST /template/update is a partial update: a field absent from the request is left alone, and only an explicit empty string clears a channel. The generated request struct could not express that. TemplateUpdateRequest's 14 channel-content fields, description and team_id were value types with omitempty, so "" encoded identically to "unset" — clearing a channel through the SDK was impossible, and TeamID: 0 could never move a template to account scope.

This marks those properties nullable in the vendored spec and regenerates. They become pointers:

// leave every other channel alone, rewrite one
_, err := c.NotificationTemplates.WriteUpdate(ctx, &flashduty.TemplateUpdateRequest{
    TemplateID:   id,
    TemplateName: name,               // still required on every update
    Feishu:       flashduty.String(src),
})

// clear a channel
Zoom: flashduty.String(""),

Why the spec is edited in place

make sync-spec pulls the whole upstream document, and the vendored copy is a few endpoints behind, so a wholesale resync would bury this schema in unrelated churn. The edit here is byte-identical to the corresponding region of the upstream spec (verified field by field), and the nightly sync job reconciles the rest on its own schedule.

Ordering

Merge flashcatcloud/flashduty-docs#303 first. It carries the same change to the source-of-truth spec; landing this one first means the next nightly sync opens a PR that undoes it.

The pointer fields are correct against the server both before and after the corresponding server-side change ships — a nil field is omitted exactly as omitempty omitted a zero value today — so this can be released ahead of it.

Checks

• make generate reproduces the committed output (39 services, 336 operations, 707 schemas); the only non-spec diff is models_gen.go + the doc comment on WriteUpdate.
• go build ./..., go build -tags e2e ./..., go vet ./..., go test ./... — all green.
• golangci-lint run (v2.11.4, the pinned version) — 0 issues.
• Both new guards mutation-verified: reverting a channel field to a value type reddens the "clears a channel" case, and dropping an omitempty reddens the "carries only the channel it changes" case.