fix(sync): skip records that fail to decrypt or decode instead of failing the whole store

[ellie]

Previously, we chose to fail loudly when there was a key mismatch in a users store. However, in the past, we did not check the key at login, so users could easily end up in a state where they inserted an invalid key, and then broke their store.

This PR ensures that even with a partially invalid store, the user can still sync their data.

This does not mean that users will be able to introduce new invalid keys, and simply makes things better for older users (pre key-checking). We still need to figure out a more straightforward way for a user to repair their entire record store if they have somehow mixed up keys, other than the purge + push/pull --force steps.

Checks

• I am happy for maintainers to push small adjustments to this PR, to speed up the review cycle
• I have checked that there are no existing pull requests for the same thing

[ellie]

@greptile-apps plz sir may i have one review 🙏

[greptile-apps]

Greptile Summary

Reviewed PR #3569 which converts hard decrypt/decode failures across all record stores into skip-and-warn behavior. Found one P1 issue: sync::build() uses eprintln! for store-level failures while the rest of the PR uses tracing::warn!, and this function can be called during interactive shell sessions.

Confidence Score: 4/5

Safe to merge; the only concern is that build-level failures in sync.rs still use eprintln! and can surface in interactive sessions.

The store-level changes are correct and well-tested. The one gap is sync.rs, which uses eprintln! for build failures while the same function is called from interactive history sessions.

crates/atuin/src/sync.rs — eprintln! should be tracing::warn! to match the rest of the PR and avoid terminal pollution in interactive sessions.

Important Files Changed

Filename	Overview
crates/atuin-client/src/history/store.rs	Converts decrypt/decode failures from hard errors to skips with tracing::warn; adds test for corrupt-record resilience; incremental_build follows the same pattern correctly using the outer-loop id variable.
crates/atuin/src/sync.rs	build() now swallows per-store failures and logs them with eprintln!, but individual-record errors inside each store use tracing::warn — inconsistent, and eprintln! can surface in interactive sessions triggered from history.rs.
crates/atuin-dotfiles/src/store.rs	aliases() now skips corrupt records with tracing::warn; test added mirrors the pattern in history store.
crates/atuin-dotfiles/src/store/var.rs	vars() applies same skip-and-warn pattern; consistent with alias store changes.
crates/atuin-kv/src/store.rs	KV store build now skips undecryptable records; skipping a newer Delete record leaves an older Create visible, but this is the accepted trade-off for a mixed-key store.
crates/atuin-scripts/src/store.rs	scripts() applies the same pattern; consistent with other stores.

_{Reviews (2): Last reviewed commit: "review fixes" | Re-trigger Greptile}

[BinaryMuse]

This seems reasonable. Do we have a good way to surface to the user that something went wrong other than them looking in the logs?

[davidolrik]

We still need to figure out a more straightforward way for a user to repair their entire record store if they have somehow mixed up keys, other than the purge + push/pull --force steps.

@ellie: I have a solution for this in #3429, which adds atuin store repair. - It has just been rebased on top of main