Querying unsymbolized profiles with maxNodes !=0 returns truncated/invalid data

```
Period: 0
Samples:
cpu/nanoseconds
1000000000: 1 2
1000000000: 3 4
1000000000: 5 6
Locations
1: 0x0 M=1 foo :0:0 s=0()
2: 0x0 M=1 bar :0:0 s=0()
3: 0xcafe03000 M=2
4: 0xcafe04000 M=2
5: 0xcafe05000 M=2
6: 0xcafe06000 M=2
Mappings
1: 0x0/0x0/0x0
2: 0x0/0x0/0x0 library.so
```

gprofile go code

```golang
gp := &gprofile.Profile{}
gp.Mapping = []*gprofile.Mapping{
{
ID: 1,
},
{
ID: 2,
File: "library.so",
},
}
gp.Function = []*gprofile.Function{
{
ID: 1,
Name: "foo",
},
{
ID: 2,
Name: "bar",
},
}
gp.Location = []*gprofile.Location{
{
ID: 1,
Mapping: gp.Mapping[0],
Line: []gprofile.Line{
{
Function: gp.Function[0],
},
},
},
{
ID: 2,
Mapping: gp.Mapping[0],
Line: []gprofile.Line{
{
Function: gp.Function[1],
},
},
},
{ID: 3, Mapping: gp.Mapping[1], Address: 0xcafe0_3000},
{ID: 4, Mapping: gp.Mapping[1], Address: 0xcafe0_4000},
{ID: 5, Mapping: gp.Mapping[1], Address: 0xcafe0_5000},
{ID: 6, Mapping: gp.Mapping[1], Address: 0xcafe0_6000},
}
gp.Sample = []*gprofile.Sample{
{
Location: []*gprofile.Location{gp.Location[0], gp.Location[1]}, // foo ; bar
Value: []int64{1_000_000_000},
},
{
Location: []*gprofile.Location{gp.Location[2], gp.Location[3]}, // 0xcafe0_3000 ; 0xcafe0_4000
Value: []int64{1_000_000_000},
},
{
Location: []*gprofile.Location{gp.Location[4], gp.Location[5]}, // 0xcafe0_5000 ; 0xcafe0_6000
Value: []int64{1_000_000_000},
},
}
gp.SampleType = []*gprofile.ValueType{
{
Type: "cpu",
Unit: "nanoseconds",
},
}
```

When we query a tree back, the symbolizer converts the query to pprof type and with default 8192 max nodes receives the following
```
PeriodType:
Period: 1000000000
Time: 2026-01-26 17:27:43.357 +0700 +07
Samples:
cpu/nanoseconds[dflt]
1000000000: 1 2
2000000000: 3
Locations
1: 0x0 M=1 foo :0:0 s=0()
2: 0x0 M=1 bar :0:0 s=0()
3: 0x0 M=1 other :0:0 s=0
Mappings
1: 0x0/0x0/0x0
```

And If I query pprof with maxNodes=0 (hitting pprof export button) I get a proper profile back.

```
PeriodType:
Period: 1000000000
Time: 2026-01-26 17:31:51.29 +0700 +07
Samples:
cpu/nanoseconds[dflt]
1000000000: 1 2
1000000000: 3 4
1000000000: 5 6
Locations
1: 0x0 M=1 foo :0:0 s=0()
2: 0x0 M=1 bar :0:0 s=0()
3: 0xcafe05000 M=2
4: 0xcafe06000 M=2
5: 0xcafe03000 M=2
6: 0xcafe04000 M=2
Mappings
1: 0x0/0x0/0x0
2: 0x0/0x0/0x0 library.so
```

From a quick look I think the truncation happens here
https://github.com/grafana/pyroscope/blob/35094213f1bfbb8ac0659168caa96bf0d03d845d/pkg/phlaredb/symdb/resolver_pprof_tree.go#L70

If a stacktrace is completely unsymbolized the samples are all merged together. If a stacktrace is partially symbolized - the tree is likely built only on the symbolized frames which is also incorrect.