Repro scripts admit vector(n,float16) and declare each parameter once (#4613)

[erikdarlingdata]

Fixes #4613
Refs erikdarlingdata/PerformanceStudio#592

Why

A Query Store plan's generated repro script left out SQL Server 2025's half-precision vector(…,float16) parameters, because the data-type check the previous fix (#4567) installed didn't admit a two-part (n,name) suffix. Separately, a batch whose statements share a parameter — including auto-parameterized @0/@1 names — declared that parameter once per statement and failed with "The variable name '@id' has already been declared" or "Must declare the scalar variable @1".

What changes

Ports erikdarlingdata/PerformanceStudio#592 (merged 6e52102) into PerformanceMonitor.PlanAnalysis/ReproScriptBuilder.cs:

• IsValidDataType: the suffix now accepts (n), (max), (p,s), or (n,name) — the last form is what admits vector(3,float16).
• Parameters are grouped by name (case-insensitively) before declaring: each name is declared once. A name that different statements type differently is left out of the declaration list, with a warning naming it.
• When the statements that share a parameter carry different compiled values, the script uses the first usable one and warns that the values disagreed.
• The "no parameters" placeholder comment now distinguishes an unparameterized query from one whose parameters all failed to declare (/* No parameters declared: see the warnings above */).
• IsSafeLiteral's numeric-literal check switched from \d to [0-9], matching PerformanceStudio, so a compiled value using another script's digits (which \d also matches but T-SQL doesn't read as numbers) still becomes a ? placeholder.

Hunk-by-hunk

1. validParameters/grouping/conflictingNames/declarableGroups/safeParameters/differingValueNames block — ported verbatim (adapted variable names already matched).
2. droppedCount now measured against validParameters, plus the two new warnings (conflicting types, differing values) — ported.
3. The "no parameters" placeholder comment split into two cases — ported.
4. IsValidParameterName's \A…\z anchors and doc comment — already in PM from Plan analysis: the repro script matches PerformanceStudio's (#4564) #4567; not reapplied (would be a no-op).
5. IsValidDataType's new suffix grammar (adds the (n,name) alternative) — ported.
6. IsSafeLiteral's three regexes switched to \A…\z — already in PM from Plan analysis: the repro script matches PerformanceStudio's (#4564) #4567. Only the \d → [0-9] change in the numeric branch was new; ported that one line.
7. Test file (+227 lines) — ported as a new class (see below); PS's ParseScript call in each test isn't present in PM (no T-SQL parser dependency in this project), so it's omitted; every Assert.Contains/DoesNotContain line is kept.

Everything else in PS's diff was already on PM's side from #4567 (the \A/\z anchor switch, CommentSafe, no USE line on control characters, the leading-digit parameter name pattern).

Consumers

Lite (Lite/Analysis) and the Darling Viewer both build against the shared ReproScriptBuilder and needed no changes; their builds are 0 errors / 0 warnings with this change.

Test plan

New class Darling/Darling.Tests/PlanSync4613ReproScriptTests.cs ports PerformanceStudio's new tests (adapted: no ParseScript helper).

RED on origin/dev (4db597907, worktree at that commit): 8 of 28 tests in the new class failed, including the vector(3,float16) type-kept pin and the parameter-dedup pins. All 8 compiled fine on dev (no new members needed) — a genuine runtime RED, not a compile failure.

Mutation: reverted IsValidDataType to dev's pattern (no (n,name) suffix) against the otherwise-fixed code — vector(3,float16) and 3 other malformed-type pins went RED (4 of 28), confirming the pin is load-bearing. Reverted back; rebuilt green.

GREEN after the full port:

Darling.Tests  Total: 151, Errors: 0, Failed: 0, Skipped: 0, Not Run: 0, Time: 4.346s

(classes run: PlanSync4613ReproScriptTests, PlanSync4564ReproScriptTests, ReproScriptBuilderHardeningTests, DocCommentHygieneTests, CommentFilterAdoptionTests)

Builds, 0 warnings / 0 errors: Darling.Tests, Lite.Tests, PerformanceMonitor.Darling.Viewer.

Lite.Tests can't run in-process on macOS (WindowsBase discovery); no Lite test touches ReproScriptBuilder directly, so none were run or skipped for this change.

CHANGELOG

SECTION: Fixed
ENTRY: - The repro script declares each parameter once ([#4626]) - When a batch's plan lists the same parameter, or an auto-parameterized @0/@1, for more than one statement, the repro script declared it once per statement and failed to run. Each name is now declared once; a name the statements give different types is left out with a warning, and a warning says when the statements' compiled values differ. Matches PerformanceStudio.
REF: [#4626]: #4626