Percentile Calculator returns 500 for stations that have no pre-calculated data file

## Summary

`POST /api/percentiles/` returns a 500 when a requested station has no pre-calculated data file. Because codes are processed as a batch, one such station also fails every other station selected alongside it.

## Steps to reproduce

Breaks (VANCOUVER AP, code 69):
https://psu.nrs.gov.bc.ca/percentile-calculator?codes=69&timeRange=10

Works (codes 213 and 267):
https://psu.nrs.gov.bc.ca/percentile-calculator?codes=213,267&timeRange=10

Adding the unsupported station breaks the working pair:
https://psu.nrs.gov.bc.ca/percentile-calculator?codes=213,267,69&timeRange=10

Selecting only stations that have data works:


Adding VANCOUVER AP (EC) (69) to that same pair fails the whole request:


Against production:

```
POST /api/percentiles/ {"stations":[213], ...} -> 200
POST /api/percentiles/ {"stations":[267], ...} -> 200
POST /api/percentiles/ {"stations":[213,267], ...} -> 200
POST /api/percentiles/ {"stations":[1142], ...} -> 500
POST /api/percentiles/ {"stations":[1142,213,267], ...} -> 500
```

The page shows "Error occurred (while getting the calculation result)."

## Cause

`app/percentile.py` builds a filename per station code and calls `StationSummary.parse_file(filename)` with no existence check, so a missing file raises `FileNotFoundError` and surfaces as a 500.

The same function already returns a clean 400 for its other two failure modes (empty station list, unsupported year range). A missing data file looks like the case that was not covered.

## Expected

A 400 naming the unsupported codes, consistent with the existing 400s, and without failing the other stations in the request.

There is precedent that a data-less station is not meant to error: ten currently-selectable stations have null pre-calculated values and return 200, rendering as "Not available" with a "Data only available for 0 of 10 years" notice, e.g.

https://psu.nrs.gov.bc.ca/percentile-calculator?codes=321&timeRange=10


The only difference between that and a 500 is whether a file exists on disk:

```
321 (file present, values null) -> 200, "Not available"
1142 (no file) -> 500
```

Am I missing something here? If the 500 is expected for some reason I haven't spotted, happy to be corrected.

PR to follow.

Start by reproducing the request against POST /api/percentiles/ and read app/percentile.py, especially the existing 400 responses and the StationSummary.parse_file(filename) call. Done means a missing-data station produces a 400 naming the unsupported code while valid stations in the same batch still succeed.

Written by the indexing model from the issue text.