Fix Prometheus exporter to sanitize malformed unit strings

[Nik-Reddy]

Description

Add SanitizeUnitName() to strip invalid characters from metric unit strings before they are used in Prometheus metric names. Units like '# RU' (from Azure Cosmos DB) previously produced invalid metric names containing comment markers and spaces.

The sanitizer replaces non-alphanumeric characters with underscores, collapses consecutive underscores, and strips leading/trailing underscores. Fully invalid units (e.g. '#') result in an empty string so no unit suffix is appended.

Fixes #6187

The Prometheus exporter does not sanitize unit strings after processing in
GetUnit(). When upstream libraries set malformed units (e.g., Azure Cosmos DB
uses # RU), the resulting Prometheus metric name contains invalid characters
like # and spaces, which breaks Prometheus scraping because # is interpreted
as a comment marker.

Expected behavior

Malformed unit strings should be sanitized so the final Prometheus metric name
only contains valid characters ([a-zA-Z0-9_]).

• # RU → metric suffix becomes _RU
• # → treated as unitless (no suffix)
• #_R.U. → metric suffix becomes _R_U

Actual behavior

The unit string # RU is appended as-is, producing metric names like
azure_cosmosdb_client_operation_request_charge_# RU_bucket which Prometheus
cannot scrape.

Steps to reproduce

1. Create a Meter with unit set to # RU
2. Configure OpenTelemetry with the Prometheus exporter
3. Query the Prometheus metrics endpoint
4. Observe the metric name contains # and spaces

Environment

• OpenTelemetry .NET SDK
• .NET 8.0
• Package: OpenTelemetry.Exporter.Prometheus.HttpListener

Changes

• Added SanitizeUnitName() method to PrometheusMetric.cs that replaces
  invalid characters with _, collapses consecutive underscores, and strips
  leading/trailing underscores
• Wired it as the final step in GetUnit() before returning
• Added 9 new test cases for malformed unit scenarios
• Updated 1 existing test expectation for multi-slash units

Type of change

• Bug fix (non-breaking change which fixes an issue)

How Has This Been Tested?

All 188 tests pass via dotnet test.

Does This PR Require a Contrib Repo Change?

• No

Merge Requirement Checklist:

• CONTRIBUTING guidelines followed
  (license requirements, nullable enabled, static analysis, etc.)
• Unit tests added/updated
• Appropriate CHANGELOG.md files updated for non-trivial changes
• Changes in public API reviewed (if applicable)

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: Nik-Reddy / name: Nik-Reddy (f8dbc36)

[martincostello]

Thanks for the PR, but this appears to duplicate the work already done in #7033?

[Nik-Reddy]

Thank you for pointing this out. I wasn't aware of PR #7033 when I submitted this. After reviewing it, both PRs take the same
approach of sanitizing the unit string in GetUnit() before it's returned.

I'm happy to close this PR in favor of #7033 since it was submitted first and already has maintainer review.
I do notice one difference: my implementation also strips trailing underscores (e.g., "##/RU!" → "RU" vs "RU_" in #7033), which might be worth incorporating there.

I'll keep a closer eye on open PRs in the future. Happy to contribute on other open issues!

[Kielek]

@Nik-Reddy, thanks for your contribution. I have copied your comment to the other PR.
We are looking always for the contribution to review existing PRs if you are interested in helping in this way.