Issues caused by different duration rounding between transactions and children spans

[cjcormack]

Some requests in our microservices are taking < 1ms. Transactions for these appear to have issues as the rounding on timestamps is at the millisecond level, rather than sub-millisecond which is used by spans. For example, in the following screenshot the transaction has rounded down to taking 0ms, rather than the fraction of a millisecond. A secondary effect of this is that graphs that show the p50, for example, show 0.00ms which is particularly unhelpful when looking at graphs.

The numbers in the JSON representation of the transaction is correct, e.g.:

"spans": [
  {
    "timestamp": 1750176119.291527,
    "start_timestamp": 1750176119.291111,
    "exclusive_time": 0.071916,

This is possibly a regression as it feels like the same issue described in #79406, which was closed as 'completed'.

[getsantry]

Assigning to @getsentry/support for routing ⏲️

[getsantry]

Routing to @getsentry/product-owners-performance for triage ⏲️