The profile header's banner, avatar, and DID drifted to different
positions on different devices. SliverAppBar.expandedHeight was
hardcoded to 220 while the header content started below a SafeArea whose
top inset varies per device (~20pt on older iPhones, ~59pt on Dynamic
Island models), so the total height and the content's starting point
disagreed: the identity block landed lower and was clipped by a
different amount on every screen.

Rebuild the header around explicit Positioned layers whose geometry is
derived from the device inset and text scale, so the banner's bottom
edge always sits a fixed distance below the collapsed app bar and the
avatar always straddles that edge by exactly half its height.

Note that SliverAppBar computes maxExtent as topPadding + expandedHeight,
so expandedHeightFor() deliberately excludes the status-bar inset —
including it counts the inset twice and makes the banner grow at twice
the inset's rate.

Changes:
- Derive header geometry from MediaQuery insets and text scale via
ProfileHeader.expandedHeightFor / maxExtentFor / collapsedExtentFor
- Replace the SafeArea + UnconstrainedBox flow layout with explicit
Positioned layers: banner, scrim, avatar, identity
- Move bio, stats, and join date into a new ProfileDetails sliver so
they are never clipped by the collapsing header, dropping the 2-line
bio truncation that the clipping required
- Set the banner overhang to avatarSize / 2 so the avatar cannot rise
above the collapsed app bar and show through the frosted overlay
- Compute collapse progress from the sliver's real max extent, so the
frosted blur no longer stalls for the first topPadding px of scroll
- Replace the per-platform iOS/Android gradient hacks with one scrim
- Make the DID tappable to copy, since it ellipsizes on narrow screens
- Add profile_header_test.dart pinning the geometry at insets of
0/20/44/59pt; 3 of its 5 tests fail against the previous layout

Verified with flutter analyze, the full test suite (948 passing), and
pixel measurements on the Android emulator at two status-bar insets:
the banner edge now tracks the inset at 1x (63px between a 24dp and
48dp inset) rather than 2x (126px).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Pre-push review of the fastlane work. No secret was ever committed, but
two things were worth fixing before this goes out to the world.

The .env and *.p8 ignore rules matched only the two exact paths in use.
fastlane also loads .env.default and .env.<environment>, and RELEASING.md
tells you to download a .p8 that is only ignored if it happens to land in
ios/fastlane/. Add repo-root patterns covering the whole class, with
.env.example kept visible.

RELEASING.md also read as if everyone holds the signing keys, which is
true for exactly one person. Say up front that it is maintainer-only and
that a fork needs its own accounts and identity, so a contributor does
not stall on step 1. Recommend the three narrow Play permissions ahead of
Admin rather than after it -- a public doc gets copied.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Play version codes are global to the app; App Store build numbers are
scoped to the version string. The distinction decides whether a number
can be reused, and the stores had already drifted (Play 1.0.5+7, App
Store 1.0.6+8), so neither can be inferred from the other or from
pubspec.yaml.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

A bare build number cannot decide the next one on iOS. Apple scopes
CFBundleVersion uniqueness to the CFBundleShortVersionString train, so
"build 8 exists" only blocks a release that reuses the same version
string -- and the lane previously reported the number with no way to
tell which train it belonged to.

Surfaces that App Store is live on 1.0.6+8 while Play production is on
1.0.5+7: the two platforms had already drifted apart, and pubspec.yaml
was behind both.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Encodes the release procedure so the next one does not rediscover it,
in particular the store-truth version check: pubspec.yaml sits behind
the stores whenever a build is uploaded from a working copy without the
bump being committed, which nearly produced a rejected upload this time.

Also records the failure modes hit while wiring fastlane up -- the
CocoaPods/Ruby env leak, the ASC account-agreement block, Play
permission propagation, and the flat_map crash on empty tracks -- since
each one presents as something other than what it is.

Guardrails match the lanes: upload as draft, never roll out or submit.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Mirrors the Play track queries on the iOS side. Android shipped a build
whose bump was never committed, leaving pubspec.yaml behind the store
and very nearly producing a rejected upload; the same trap exists on
App Store Connect, so make it equally cheap to check there.

Swallows the not-found error rather than raising, since "no build in
that state" is a legitimate answer, and logs the underlying message at
verbose level so a real failure is still diagnosable -- which already
paid off: it surfaced an expired App Store Connect agreement that
presents as an empty result.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

fastlane auto-loads fastlane/.env, which is a much better home for the
App Store Connect key details than shell exports that vanish with the
terminal. Ignore it on both platforms before anything lands in it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Querying Play with the new service account showed production live on
1.0.5+7 while the committed pubspec said 1.0.5+6 -- a build was uploaded
straight from a working copy without the bump ever being committed, so
git understated the real version. Version code 7 was therefore already
taken and the 1.1.0+7 artifacts would have been rejected on upload.

Other tracks are well behind (alpha 4, internal 2), so 8 clears
everything live.

Document the pre-bump check in RELEASING.md so the next release starts
from what the store reports rather than from pubspec.yaml.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>