fix(workflow): let a webhook body fill a built-in trigger field it names

[Chessing234]

Fixes #5769.

TriggerContext::get_field matches the built-in names before it consults webhook_fields, so a webhook body key called text — or author, timestamp, emoji, message_id — resolves to the built-in's empty default. A webhook trigger populates none of those, so {{trigger.text}} renders empty, the run still succeeds, and the posted value simply vanishes from the message. Renaming the key to anything non-reserved works, which is what makes it look like a template bug rather than a shadowing one.

The fall-back to webhook_fields happens only when the built-in is empty, so a body still cannot spoof a field the trigger actually set — a message_posted trigger's author stays authoritative even if the payload names one. There is a test for exactly that.

evaluate_condition had the same split: it registered the webhook fields as trigger_<key> and then overwrote them with the built-ins, including the empty ones, so a condition on trigger_text compared against "" while a template on the same field would now read the body. Both go through get_field now, so the two agree.

Related: #3503 fixes the manual-trigger half of this in command_executor.rs by mapping known --inputs keys onto the built-in fields. This is the same root cause a layer down — with get_field fixed, the manual path resolves too, without the caller-side mapping.

Verified locally: cargo test -p buzz-workflow --lib (160 passed, 5 of them new), cargo clippy -p buzz-workflow --all-targets clean, cargo fmt --all --check clean. I did not run a webhook end to end against a relay — the two new relay-facing claims (the body lands in webhook_fields, channel_id is the only built-in a webhook populates) are read from bridge.rs:1879-1894.

[themiguelamador]

No findings. I traced every TriggerContext producer and verified the precedence change across event, manual, scheduled, and webhook runs. The linked issue's text reproduction is fixed for both templates and condition evaluation, while populated server-derived fields such as author and channel_id still take precedence over same-named payload fields.

Verified with all 160 runnable buzz-workflow library tests (2 Postgres-only tests ignored by the crate), strict Clippy, rustfmt, and git diff --check.