Reduce NEWS.md merge conflicts between parallel PRs (#1509)

Co-authored-by: sbfnk <sebastian.funk@lshtm.ac.uk>

Author: sbfnk-bot

.gitattributes

@@ -0,0 +1,3 @@
+# Bullets are independent lines, so keep both sides instead of conflicting
+# when parallel PRs each add a news item.
+NEWS.md merge=union

CLAUDE.md

@@ -87,7 +87,14 @@ test_that("calc_CrI works as expected with default arguments", {
 - Don't reference issue numbers - those are for PRs, not NEWS
 - Don't mention code generators/reviewers or @ mentions
 - Keep entries relatively short (default: 1 sentence, but can be more if needed)
-- Organize by section: "Package changes", "Model changes", "Bug fixes", "Documentation"
+- Organise by section: "New features", "Breaking changes", "Model changes",
+  "Package changes", "Bug fixes", "Documentation"
+- Add your bullet to a section that already exists rather than creating a
+  heading. The development section is seeded with all standard headings,
+  empty, when the version is bumped, and the release PR removes any that are
+  still empty. This keeps parallel PRs from colliding on the headings
+  themselves, and lets `NEWS.md merge=union` in `.gitattributes` resolve the
+  bullets automatically
 - Start with action verbs: "Added", "Fixed", "Updated", "Changed"
 - Bug fixes often start: "A bug was fixed where..."
 

NEWS.md

@@ -1,19 +1,24 @@
 # EpiNow2 (development version)
 
-## Bug fixes
+## New features
 
-- A bug was fixed where disabling the weekly reporting effect produced a spurious convergence warning from a degenerate day-of-week simplex.
-- A bug was fixed where `estimate_infections()` could emit a spurious "the largest R-hat is NA" convergence warning caused by deterministic delay PMFs being monitored; these are no longer monitored.
+## Breaking changes
+
+## Model changes
 
 ## Package changes
 
 - Increased the default number of warmup iterations in `stan_sampling_opts()` from 250 to 500 to reduce intermittent non-convergence of individual chains.
 
 ## Bug fixes
 
+- A bug was fixed where disabling the weekly reporting effect produced a spurious convergence warning from a degenerate day-of-week simplex.
+- A bug was fixed where `estimate_infections()` could emit a spurious "the largest R-hat is NA" convergence warning caused by deterministic delay PMFs being monitored; these are no longer monitored.
 - A bug was fixed where the prior on the initial reproduction number was applied with an incorrect Jacobian, shifting it upwards by a factor of `exp(sdlog^2)`. A `LogNormal(mean = 2, sd = 1)` prior was applied as though it had a mean of 2.5.
 - A bug was fixed where the mean reproduction number over the observation window was left uninitialised, so chains started from a value drawn across the whole of `exp(-2)` to `exp(2)` rather than from the user's Rt prior.
 
+## Documentation
+
 ## Internal
 
 - Removed redundant namespace qualification (`pkg::fn`) on imported functions across the package, and excluded `data-raw` from linting, so `lintr` passes cleanly.